What's New ── Update Log── Daily new articles, revisions, and published videos

A chronological record of Advertising Regulation in Japan (Promotional Material Review) and Artificial Intelligence site updates. New article publications, revisions to existing articles, and video releases — all gathered here, newest first.

November 4, 2026New

AI and Emerging Technology

Paper review published — Do the Gains of Self-Improving Agents Hold Up Outside Training Conditions?

Are the gains of agents that rewrite their own pipelines real capability, or fitting to the test scenarios? SelfImprove-Bench, a preprint not yet peer-reviewed, grades transfer on a G0 to G4 ladder and reports that gains of existing RSI methods were inconsistent across levels. This review separates what the abstract shows from what it does not. Published in both Japanese and English.
AI and Emerging Technology

Paper review published — The Task Passed, but Unrelated Files Are Gone

Coding agents with shell access can finish a task while deleting unrelated files or weakening permissions. VeriHarm, a preprint not yet peer-reviewed, runs correctness and harm tests together and reports average task accuracy of 95% but an average safe success rate of 37%. This review reads what those figures mean and where they stop. Published in both Japanese and English.
AI and Emerging Technology

Paper review published — Training Computer-Use Agents on Verified Experience

NorthCUA's technical report, a preprint not yet peer-reviewed, trains screen-operating AI agents on experience whose outcomes were checked, and reports state-of-the-art results among open-source models. This review reads the scope of that comparison and how training records differ from regulated work records. Published in both Japanese and English.

November 3, 2026

AI Literacy

AI Literacy S4-09 published — Make AI Keep the Request Ledger and Priorities with One Rule

Handing priorities to AI means setting the method once. Limit the ledger, the scoring formula and the task in progress to one each, write it where the AI reads every time, and the order stays consistent. Published in both Japanese and English.
AI and Emerging Technology

Paper review published — Can a causal graph explain why a reinforcement learning agent acted?

Reinforcement learning agents can perform well while their reasons stay hidden. The preprint CausalXRL proposes learning a causal graph to explain agent decisions at several levels. This review separates what the abstract shows from what it leaves open. Published in both Japanese and English.
AI and Emerging Technology

Paper review published — Where do AI agents that act before being asked go wrong?

Proactive agents must decide on their own whether, when and how far to help. The preprint HALO built 8,000 expert-annotated responses and reports that over-proactivity was the most common failure. This review reads it within the abstract. Published in both Japanese and English.
AI and Emerging Technology

Paper review published — Before running heavy reasoning, let a light model estimate whether it is worth it

AI progress has come largely from pouring more compute into larger models and longer reasoning. This preprint formalises screening candidates with a light predictor before heavy execution, and links the saving to the density of useful interactions. Theory and evidence are read separately. Published in both Japanese and English.

November 2, 2026

AI Literacy

AI Literacy S4-08 published — Interrupting a Running AI Task: Put 4 Things in One Message

To give a working AI a separate request, state the interruption, the scope, the fate of the current work and the way back. Requests for another product or material get their own conversation. Published in both Japanese and English.
Column

Column No. 145 published — UK Planning AI Cuts Officer Time 58%, but Reports No Figure for Decision Weeks ── The 58% measures officers' working time, so applicants' waiting weeks and decision quality have to be checked with other numbers.

Google DeepMind says its planning AI, PlanAssist, cut officers' assessment time by 58% in three UK councils. The figure does not show how long applicants wait or whether decisions are sound. A look at the 21-day consultation, appeal outcomes and the 70% of drafts passed unedited. Published in both Japanese and English.
Column

Column No. 146 published — A 4,598-Person Trial in Five Countries Measured Forgiveness in the Forgiver ── The trial reviewed here showed gains only in the forgiver, so decide on forgiving for your own sake and leave the other person's responsibility standing.

Is forgiving someone done for that person, or for yourself? A randomized trial of 4,598 people in five countries found lower resentment, depression and anxiety in the forgiver. Other studies show why amends and safety matter, and what AI errors add to the question of responsibility. Published in both Japanese and English.

November 1, 2026

AI Literacy

AI Literacy S4-07 published — Japanese First, English After Go-Live: Telling AI Once

If you keep repeating an order to an AI, write the start condition so it can be checked, make it a four-part sentence, and keep it where it loads every time. Published in both Japanese and English.
Copilot

Microsoft Copilot Know-how updated — 3 added, 0 revised (SharePoint / OneDrive 1, Copilot Chat 1, Grounding Copilot in Your Stored Content 1)

Including: Pick ready-made file prompts from Prompt Gallery in OneDrive; Keep related chats, files and instructions together in Projects; Bring life sciences sources into Copilot with the new connectors. Only tips matched against their sources are published. Published in both Japanese and English.
AI and Emerging Technology

Paper review published — Does Agent Memory Beat Simply Searching the Raw History?

The Memory Tax preprint reports that many agent memory systems fall short of naive RAG in many cases and proposes measuring gains and losses separately. What the abstract does and does not establish. Published in both Japanese and English.
AI and Emerging Technology

Paper review published — Making Professional Tasks Executable Before Training on Them

Training tasks can lack information or contradict themselves. The WorkFactory preprint plans information dependencies and repairs tasks through probe runs. How to read its Elo gains, and their limits. Published in both Japanese and English.
AI and Emerging Technology

Paper review published — Can Agents Post-Train a Model Across Several Domains at Once?

The ParetoRSI-Gym preprint holds everything but the data fixed and measures agent-led post-training across five domains. How to read its 9.2% average gain, and where the abstract stops. Published in both Japanese and English.

October 31, 2026

AI Literacy

AI Literacy S4-06 published — AI Offered Three Options. Set the Order Yourself: 1, Then 3, Then 2

A record of ordering an AI's three options as 1, 3, 2 shows how to sort options by whether they can be undone, and how to sequence AI pre-reading in material review. Published in both Japanese and English.
Copilot

Microsoft Copilot Know-how updated — 2 added, 0 revised (Copilot Chat 1, Operating AI Agents 1)

Including: Create, update and delete in third-party services from Copilot Chat; Review connector write/delete tools in the admin center and disable connectors. Only tips matched against their sources are published. Published in both Japanese and English.
AI and Emerging Technology

Paper review published — The file is genuine, yet the model cannot write code

Do quantized LLMs from public registries actually work? A preprint tested 305 official Ollama tags and found 7 that solve almost nothing, though the files were not corrupted. Published in both Japanese and English.
AI and Emerging Technology

Paper review published — When AI agents train a model and people grade it, what actually improves?

Can AI agents train language models and improve them from human feedback? In the RSI Arena preprint, ratings rose for eight of ten models, while benchmark gains stayed modest. Published in both Japanese and English.
AI and Emerging Technology

Paper review published — What is an agent you can hand work to, and how should it be measured?

When work is delegated to an agent, is the model all that needs checking? This preprint survey treats agents as models plus runtime harnesses and judges them by verifiable outcomes. Published in both Japanese and English.

October 30, 2026

AI Literacy

AI Literacy S4-05 published — Requests Grew from 19 to 71 Characters: What Was Added

Median requests grew from 19 to 71 characters. What was added: a definition of done and an order of work. Four parts to check, applied to material review. Published in both Japanese and English.
Copilot

Microsoft Copilot Advanced updated — 1 added, 0 revised (Power Automate × Python 1)

Including: Draft a Run Python script with Copilot natural language to script (preview). Only tips matched against their sources and passing the code checks are published. Published in both Japanese and English.
AI and Emerging Technology

Paper review published — When an AI keeps rewriting itself, how do you check that it improved?

When an AI rewrites its own prompts, memories and even its rules for improving, what shows it actually got better? This preprint survey divides recursive self-improvement into diagnosis, proposal, evaluation and update, and organises trust requirements around improvement validity, constraint preservation and effective oversight. Read within the abstract, with links to pharmaceutical change control. Published in both Japanese and English.
AI and Emerging Technology

Paper review published — Letting the main agent write the collaboration plan as a program, task by task

Instead of fixing how AI agents divide the work, the main agent writes an executable workflow for each task and runs sub-agents through it. The DynaFlow preprint trains only the main agent and argues that information flow across stages is key. Read within the abstract, with links to multi-stage document work in pharma. Published in both Japanese and English.
AI and Emerging Technology

Paper review published — Finding brief edits in long videos on separate audio and visual timelines

Edits that swap a brief moment inside an otherwise authentic long video are hard to catch with a whole-video verdict. This preprint builds AV-FABLE, a benchmark with separate audio and visual timelines, and reports that reading an AV-LLM's internal features recovers manipulation evidence that asking it in words does not. Read within the abstract, with links to video governance in pharma. Published in both Japanese and English.

October 29, 2026

AI Literacy

AI Literacy S4-04 published — Have AI List the Items Before Building a Large Section

Before AI builds a large section, have it list the items in a four-column table and agree on it. Scope errors get fixed early, and the list guides design and the final check. Published in both Japanese and English.
Copilot

Microsoft Copilot Advanced updated — 1 added, 0 revised (Teams × Python 1)

Including: Check the 28 KB limit before posting to a Workflows webhook and back off when throttled. Only tips matched against their sources and passing the code checks are published. Published in both Japanese and English.
AI and Emerging Technology

Paper review published — How Far Can You Trust a Local Generative AI Where There Is No Signal?

Many IoT settings lose their network. A preprint proposes a connectivity-regime framework and signed knowledge packs, and tests them with offline RAG on a small device. A validly signed poisoned pack passed every pack check. This review reads the abstract closely and connects it to pharmaceutical document control. Published in both Japanese and English.
AI and Emerging Technology

Paper review published — Can a Web Agent's Pass/Fail Checker Be Reused When Requirements Change?

Can the program that judges a web agent's work be reused after requirements change? The AgentProofBench preprint tests frozen acceptance rules against changed requirements and real execution. Verifier errors and broken deliveries appeared together. This review reads the abstract and links it to change control and revalidation. Published in both Japanese and English.
AI and Emerging Technology

Paper review published — Why Do AI Answers Get Longer and Less Accurate as They Satisfy the Rubric?

Rubric-based RL rewards coverage, so trained models drift toward verbose answers with more errors. A preprint proposes the FORK reward, which also requires each unit of a response to be necessary and factually correct. This review reads its medical benchmark results within the abstract and links them to promotional material review. Published in both Japanese and English.

October 28, 2026

AI Literacy

AI Literacy S4-03 published — Have the Previous AI Write the Hand-off Note

To move work to another AI, have the AI that holds the conversation write a four-part hand-off note, read and correct it, and have the receiver play it back before starting. Published in both Japanese and English.
Copilot

Microsoft Copilot Know-how updated — 5 added, 2 revised (PowerPoint 1, OneNote 1, Outlook 1, Excel 2, Copilot Chat 1, SharePoint / OneDrive 1)

Including: Attach an existing deck so a new presentation follows its theme and styles; Capture audio, photos and typed notes in one session on Android; Open Copilot settings in classic Outlook for Windows. Only tips matched against their sources are published. Published in both Japanese and English.
Column

Column No. 142 published — Springer Nature's Review AI Keeps Manuscripts In-House and Judgment Human ── The review AI removes the confidentiality barrier to using AI on manuscripts; checking its flags, disclosing its use and deciding on the paper remain human work

On 26 October 2026 Springer Nature and Google DeepMind announced an AI prototype that reads manuscripts and flags passages for reviewers, running only inside the publisher's own environment. It solves confidentiality; checking flags, disclosing AI use and deciding on papers stay with reviewers and editors. Published in both Japanese and English.
Column

Column No. 143 published — AI Chats Are Discoverable: Court in Heppner Denies Privilege ── What employees type into AI about a dispute can be produced in U.S. litigation, and records made outside a lawyer's direction carry no privilege

Law.com reported on 26 October 2026 that companies face discovery requests for generative AI chats and logs. Rule 34 reaches data in any medium, and in Heppner a court denied privilege to records made with Claude. What shapes how records are treated, tool terms, stoppable deletion and counsel's direction, must be settled before typing. Published in both Japanese and English.
Column

Column No. 144 published — "I Just Got Lucky": 62 Studies Put the Impostor Habit at 9–82% ── The habit comes from locating success outside oneself; working with AI adds the opposite error, so the check is a record of what one actually did

Crediting success to luck has been measured as part of the impostor phenomenon since Clance and Imes named it in 1978; 62 studies put it at 9–82%. Aalto University found ChatGPT users overrated their scores. Since we misjudge our work both ways, record your own decisions apart from the AI output. Published in both Japanese and English.

October 27, 2026

AI Literacy

AI Literacy S4-02 published — Give AI the Goal and the Limits, and Let It Staff the Team

Staffing an AI team can be delegated. Write the goal, finish line, limits and stop points, review the staffing plan first, and ask for a report on who did what and what was missed. Published in both Japanese and English.
Copilot

Microsoft Copilot Know-how updated — 1 added, 0 revised (Operating AI Agents 1)

Including: Query Power Platform assets in natural language with the Cowork admin plugin. Only tips matched against their sources are published. Published in both Japanese and English.

October 17, 2026

AI Literacy

AI Literacy S4-01 published — Queue New Ideas for AI as Reserved Work

To hand an AI the next request without stopping it, open with "reserved work", state content, done condition and start condition in one message, and keep it on a list. Published in both Japanese and English.
Copilot

Microsoft Copilot Know-how updated — 1 added, 0 revised (Operating AI Agents 1)

Including: Give Dataverse agents least-privileged roles via Entra agent identity. Only tips matched against their sources are published. Published in both Japanese and English.
AI and Emerging Technology

Paper review published — Does the care an LLM recommends cost more than the treating physician's?

A peer-reviewed NEJM AI article prices LLM-written and physician care plans on the US Medicare fee schedule. The authors report higher AI costs, mostly from added tests, and a gap cost-aware prompting did not close. Published in both Japanese and English.
AI and Emerging Technology

Paper review published — Converting existing language models to read bytes

A peer-reviewed Nature paper converts existing subword language models into byte-level models via a two-stage procedure with minimal extra training. We read what the abstract claims, its limits, how it was covered, and links to sequence and document work in pharma. Published in both Japanese and English.
AI and Emerging Technology

Paper review published — When an agent keeps rewriting its own scaffolding, what justifies accepting a change?

An arXiv preprint organizes harness self-evolution, revising prompts, memory, tools and code with fixed weights, by separating update targets from proposal and admission, and derives an admission rule using independent validation data. We read its survey nature, limits and links to change control. Published in both Japanese and English.

October 16, 2026

AI Literacy

AI Literacy S3-10 published — What the 14 vague-word requests to AI turned into

Most of 14 requests with vague words became measurable within one or two turns. A list of the restatements, and four steps for the second message. Published in both Japanese and English.
Copilot

Microsoft Copilot Know-how updated — 1 added, 0 revised (SharePoint / OneDrive 1)

Including: Add a Copilot-curated knowledge base to a SharePoint site. Only tips matched against their sources are published. Published in both Japanese and English.
AI and Emerging Technology

Paper review published — Teaching a chest X-ray agent to separate when to seek evidence from which measurements to trust

X-Ragent trains a chest X-ray agent to decide when auxiliary evidence is needed and to block inapplicable measurements through a validity gate, with process rewards applied only to correct diagnoses. What the abstract shows, where it stops, and how it connects to pharmaceutical and regulatory practice. Based on a preprint. Published in both Japanese and English.
AI and Emerging Technology

Paper review published — Can an agent stop and rethink when its tools return bad output?

Tools connected via MCP return noisy feedback and users correct agents mid-task. Agent-Reflex trains reflection and failure recovery in three stages. What the abstract shows, where it stops, and how it connects to pharmaceutical practice. Based on a preprint. Published in both Japanese and English.
AI and Emerging Technology

Paper review published — A teacher that knows the answer teaches the model to assert without checking

In on-policy self-distillation, a teacher that sees a verified solution drifts toward shortcut reasoning and output entropy drops. VRSD counters this with system prompts. What the abstract shows, where it stops, and how it connects to pharmaceutical practice. Based on a preprint. Published in both Japanese and English.

October 15, 2026

AI Literacy

AI Literacy S3-09 published — Have AI State the Essence of Your Feedback, Then Make It a Rule

After giving the same feedback twice, have the AI analyse the feedback itself. Write it as a one-sentence rule with what changes and why, and store it where every session reads it. Published in both Japanese and English.
Copilot

Microsoft Copilot Know-how updated — 1 added, 0 revised (Copilot Chat 1)

Including: See which memories and past chats shaped a Copilot Chat answer. Only tips matched against their sources are published. Published in both Japanese and English.
AI and Emerging Technology

Paper review published — Beyond "Was It Made by AI?": An Agent That Checks Whether the Scene in an Image Actually Happened

Knowing that an image is not AI-generated does not tell you whether the event it shows is real. This article reviews EviScene, an agent that takes only an image and a generic instruction, gathers outside evidence, and returns a supported or refuted verdict with an explanation, along with the EviLens benchmark. It is a preprint; this review separates what the abstract shows from what it leaves open. Published in both Japanese and English.
AI and Emerging Technology

Paper review published — Agents That Model What Their Partner Knows and Intends

When AI agents work with people or with other agents, they need some estimate of what the other side knows, intends, and is likely to do. This article reviews a framework for "mental-model-enabled agents" that learns such an estimate and uses it when choosing actions. It is a preprint; this review separates what the abstract shows from what it leaves open. Published in both Japanese and English.
AI and Emerging Technology

Paper review published — Choosing Among Options Without Asking a Model to Write the Answer

When a language model is asked to pick from a list by writing out its answer, the choice can change with the order of the options, and "no answer" is hard to tell apart from uncertainty. This article reviews Jeva, an architecture designed around four structural requirements of decision problems. It is a preprint; this review separates what the abstract shows from what it leaves open. Published in both Japanese and English.
Column

Column No. 139 published — Utah's Health AI Gets Six Outside Evaluators, but No Rule on What AI Must Not Decide ── Outside evaluators test the developers' claims; what the AI may not decide is written by the state, agreement by agreement

On October 5, 2026, Utah widened its medical AI sandbox and named six outside evaluators. They test whether AI does what developers claim; limits such as no new prescriptions or dose changes are written by the state in each agreement, and no general rule exists yet. Published in both Japanese and English.
Column

Column No. 140 published — NYC's AI Hiring Law: Only 18 of 391 Employers Posted Audits ── If no one outside can check whether the required audit was done, violations go unseen and no one is held responsible

NYC requires bias audits for AI hiring tools, yet a 2024 study found only 18 of 391 employers had posted one. State auditors found 17+ possible violations in 32 audits where the city found one. Published in both Japanese and English.
Column

Column No. 141 published — Alone for 15 Minutes, 67% of Men Chose an Electric Shock ── Boredom signals that attention has nowhere satisfying to go; read it before filling it with AI

In Wilson's 2014 study, 67% of men left alone to think for 15 minutes shocked themselves. Should boredom be avoided? The evidence says read it as a signal before AI fills the gap. Published in both Japanese and English.

October 14, 2026

AI Literacy

AI Literacy S3-08 published — Ask What an AI's Word Means Before You Approve

Skim past an AI's word and you agree to the AI's meaning. Before approving, ask what it means here and restate it in your own words. Published in both Japanese and English.
AI and Emerging Technology

Paper review published — Training Long-Horizon Terminal Agents on Environments That Do Not Do the Work for Them

An explainer on TerminalHorizon, a fully automated data synthesis engine for long-horizon terminal agents. It targets a specific failure: synthesized environments that complete or simplify the exploration, implementation and verification the agent was meant to perform. This is a preprint; we separate what the abstract shows from what it leaves open. Published in both Japanese and English.
AI and Emerging Technology

Paper review published — Telling an LLM Not to Think Does Not Make It Stop

An explainer on Thinking Inertia, a study of how to measure LLMs' no-thinking behavior. Splitting each response into a pre-answer trace and a final answer, the authors find that visible inference persists under explicit no-think controls and grows as the answer space opens. This is a preprint; we separate what the abstract shows from what it leaves open. Published in both Japanese and English.
AI and Emerging Technology

Paper review published — When Coding Agents Build Agents, What They Demonstrate Is Not What They Deliver

An explainer on AgentSWE, a benchmark asking whether coding agents can build the agents their users actually need. Tasks are framed as commissions: builders iterate on illustrative cases, and acceptance runs the frozen submission on held-out cases. No evaluated agent came close to meeting the requirements. This is a preprint; we separate what the abstract shows from what it leaves open. Published in both Japanese and English.
Column

Column No. 136 published — Anthropic Report: Device Makers and ISPs, Not Labs, Can Stop AI Self-Replication ── The point where spread can be stopped lies with device defaults and ISP monitoring, not with model developers

On 12 October 2026 Anthropic reported that small AI models may soon copy themselves onto cheap IoT devices. Only GLM-5.2 completed replication, in tests that removed the break-in step. Developers cannot recall released weights, so stopping a spread falls to device defaults and ISP monitoring, while organisations must count and lock down their own devices. Published in both Japanese and English.
Column

Column No. 137 published — SEC Warns on AI Washing, Yet Fined Listed Firm Presto Nothing ── Regulators check late and rarely, so the seller must hold evidence for an AI claim in advance and the buyer must ask for it before signing

Quartz reported in October 2026 that the SEC warns about AI washing but rarely acts, and Presto, a listed company, paid no fine. The errors in sanctioned cases lay in human labour, third-party technology and missing tests, all invisible from outside. Sellers must hold the records before making a claim; buyers must ask for them in writing before signing. Published in both Japanese and English.
Column

Column No. 138 published — Praise Scores 0.12: People Grow on Information About Fixing the Task ── Whether praise helps depends less on the person praised than on whether the words say what to fix next

Teacher praise shows an effect size of only 0.12 on achievement. Across feedback research, over a third of interventions lowered performance, and praise aimed at the self pulls attention off the task. People who grow receive information on what to fix. With the rolled-back GPT-4o update in view, this column sets out how to read praise from people and from AI. Published in both Japanese and English.

October 13, 2026

AI Literacy

AI Literacy S3-07 published — Define 'Explain the Essence, Don't Read Aloud' in 4 Parts for AI

Split the complaint about reading aloud into speaker, listener, two things to add and wording not to change, then hand that to the AI. For pharmaceutical material, keep efficacy wording and numbers fixed. Published in both Japanese and English.
Copilot

Microsoft Copilot Know-how updated — 3 added, 0 revised (Excel 3)

Including: Turn workbook data into an interactive canvas with a prompt; Describe the data you need and let Copilot fetch it (get data); Have Copilot find inconsistent entries and duplicates (clean data). Only tips matched against their sources are published. Published in both Japanese and English.

October 10, 2026

AI and Emerging Technology

Paper review published — Showing People What an Agent Actually Did, With the Evidence Attached

A preprint proposes EBG, a training-free method that groups source-linked evidence from long agent runs into a behavior graph, and AgentMonBench to test it. What it claims, where the limits are, and how it connects to audit trails in pharma. Published in both Japanese and English.
AI and Emerging Technology

Paper review published — An Agent That Learns an Environment by Rewriting Its Rulebook, Not Its Weights

A preprint describes Memento 3, where a frozen LLM agent writes environment rules as a natural-language rulebook, compiles it to code, and accepts updates only after replay checks. Its claims, limits, and links to change control in pharma. Published in both Japanese and English.
AI and Emerging Technology

Paper review published — Letting Search Agents Make Small Decisions Fast, With Confidence Scores That Mean Something

A preprint proposes SearchJev, a model that scores the options in a search agent's short decisions directly instead of generating text, calibrates its confidence, and hands uncertain judgments to a larger reasoning model. Claims, limits, and pharma links. Published in both Japanese and English.

October 9, 2026

AI Literacy

AI Literacy S3-06 published — Have AI translate business needs into developer terms

Write requirements in your own business language. Have AI restate them for system developers, mark its assumptions, and check the intent before building. Published in both Japanese and English.
Copilot

Microsoft Copilot Advanced updated — 2 added, 1 revised (Authentication, Permissions and Governance 1, Power Automate × Python 1, SharePoint / OneDrive × Python 1)

Including: Send a SharePoint Online token with every upload-session chunk; Write a @workflow_tool for Azure Functions Hosted Skills workflows; Fetch thumbnails in one Graph request with a bearer token, no redirect. Only tips matched against their sources and passing the code checks are published. Published in both Japanese and English.
AI and Emerging Technology

Paper review published — When an agent rewrites its own playbook, can it keep the reasons?

LLM agents can rewrite their own procedures from experience without retraining. EVISKILL links each edit to replayable execution evidence, verifies it by targeted replay, and separates provisional retention from final adoption. A review of the preprint within the limits of its abstract, with its limitations and links to change control in pharma. Published in both Japanese and English.
AI and Emerging Technology

Paper review published — Do agents under pressure report things that are not true?

Under what conditions do LLM agents become more likely to deceive? DecepEval pairs neutral and induced versions of the same task and measures deception under four conditions drawn from fraud theory. A review of the preprint within the limits of its abstract, with its limitations and links to data integrity in pharma. Published in both Japanese and English.
AI and Emerging Technology

Paper review published — Does training models to reason more briefly make their chain of thought less trustworthy?

Does training reasoning models to use fewer tokens undermine oversight through chain of thought? This preprint measures faithfulness and monitorability separately under three efficiency methods and reports that the first tends to fall while the second holds up. A review within the limits of the abstract, with links to explainability in pharma. Published in both Japanese and English.

October 8, 2026

AI Literacy

AI Literacy S3-05 published — Give AI the reader level as a number: 90 of 100 understand

Do not ask AI for "easy to understand". Give it a reader group and the share who must understand, then test with people close to those readers. Published in both Japanese and English.
AI and Emerging Technology

Paper review published — Tool-using agents act before the evidence is in

A correct final state does not prove that a tool-using LLM agent acted on evidence it had already checked. This explainer covers a preprint that locates where the evidence-to-action chain breaks across judgment, single actions, and dependent workflows, reading only what the abstract supports, with limits and links to pharmaceutical business systems. Published in both Japanese and English.
AI and Emerging Technology

Paper review published — Can models that read context but return typed answers handle content moderation?

An explainer on a preprint that evaluates System One Models, which read natural-language context but return typed choices or probabilities, for online moderation. It covers how rules, retrieved precedents, and answer-space limits changed decisions, the limits of confidence calibration, and links to pharmaceutical material review. Published in both Japanese and English.
AI and Emerging Technology

Paper review published — Even correct memories can make agents defer to users

LLM agents with long-term memory can bend answers toward what users believed in the past. This explainer covers a preprint arguing that sycophancy arises even from correct memories, and proposing MemAdapter to calibrate memory influence with counterfactual reasoning and reflection, with limits and links to medical information work. Published in both Japanese and English.

October 7, 2026

AI Literacy

AI Literacy S3-04 published — Symbols a Newcomer Cannot Read Fail, and AI Can Fix Them

The first-time reader, not the maker, decides whether a symbol works. Have the AI list every term, fix each one, and test with a reader who lacks the context until none are left. Published in both Japanese and English.
Copilot

Microsoft Copilot Advanced updated — 2 added, 0 revised (Agents × Python 1, Power Automate × Python 1)

Including: Add an Agent to a Python Azure Function with the markdown_agent binding; Call an Agent from a Durable Functions orchestrator with context.call_agent. Only tips matched against their sources and passing the code checks are published. Published in both Japanese and English.
Copilot

Microsoft Copilot Know-how updated — 8 added, 1 revised (Copilot Chat 3, Teams 1, Notebooks / Pages 1, PowerPoint 1, Outlook 2, Grounding Copilot in Your Stored Content 1)

Including: Get an alternative answer with Try Again or Switch Model; Ask follow-up questions in chat from Copilot Search results; Find past meetings by topic or keyword in Copilot. Only tips matched against their sources are published. Published in both Japanese and English.
AI and Emerging Technology

Paper review published — How Should the Same Model Verify What a Long-Horizon Agent Produced?

When an LLM agent attempts the same long-horizon task several times, correct claims end up scattered across rollouts. This review covers VeriHarness, a preprint on deciding which claims to trust without reference answers or rubrics, its observation that consensus can hide errors, its limits, and what it means for regulated document work in pharma. Published in both Japanese and English.
AI and Emerging Technology

Paper review published — Where Does the Science Break in an AI-Written Paper?

AI-generated papers can look plausible section by section while the reasoning that ties them together falls apart. This review covers a preprint that measures that failure with six measures across structure, argument and artifacts, builds the SciSlopBench benchmark, and proposes SciSlopHarness, with its limits and its relevance to regulated pharma documents. Published in both Japanese and English.
AI and Emerging Technology

Paper review published — More Answer Options Do Not Make an AI Judge More Discriminating

Direct-decision models turn text into labels and scores and are handy for classification and automatic evaluation, but they do not necessarily use the rating scale a user supplies. This review covers a preprint reporting ordinal scale-utilization bias, where decisions crowd toward the middle and use a shrinking share of the scale as it gets finer, with its limits and what it means for rating tasks in pharma. Published in both Japanese and English.

October 6, 2026

AI Literacy

AI Literacy S3-03 published — It Is Fine to Tell AI 'I Don't Understand' Its Report

You can say you don't understand an AI report full of numbers. Have it lead with what changed, then before and after, then the numbers, and you can make the call. Published in both Japanese and English.
AI and Emerging Technology

Paper review published — Do a Research Agent's Explanations Make Experimental Forecasts Better?

AI research agents attach explanations to the experiments they plan. This preprint measures how much those explanations improve outcome forecasts by pairing forecasts that differ only in the context given, and reports that its preregistered decision on Tox21 and OpenML tasks was inconclusive. We explain the design, the results, and the limits. Published in both Japanese and English.
AI and Emerging Technology

Paper review published — Keeping Graders That Reward Missing Content Out of Reinforcement Learning

In rubric-based reinforcement learning, an LLM judge can award a high criterion score even when the response lacks the required information. This preprint names the failure "vacuous credit" and proposes MetaRubric, which gives credit only with evidence and revises rubrics from policy responses. We cover its medical benchmark results and its limits. Published in both Japanese and English.
AI and Emerging Technology

Paper review published — Rewriting Assisted Successes into Training Data That Works Without the Assistance

Successful trajectories on hard tasks are useful training data, but the specialized harnesses that produced them may not exist at deployment. This preprint proposes Recursive Self-Rewrite, which uses one base model to find solutions under several harnesses and rebuild them as trajectories under a general harness. We cover the terminal-benchmark results and the limits. Published in both Japanese and English.
Column

Column No. 124 published — Google Freezes Its OSS Bug Bounty as AI Reports Swamp Reviewers ── AI cut the cost of writing a report, not the cost of checking one; the defence is to demand reproducible proof at intake

On 1 October 2026 Google stopped accepting product vulnerability reports to its OSS VRP, saying most automated submissions were invalid. At curl, confirmed reports fell below 5%. AI made reports cheap, but checking still costs the receiver. What Google and curl changed in response. Published in both Japanese and English.
Column

Column No. 125 published — Ivo Sage Hits 91% per Criterion but Passes Every Criterion on 16% of Contract Tasks ── The 91% counts criteria, not finished contracts; it describes drafts for a lawyer to review, and re-testing stays with the user

On 1 October 2026 the legal AI company Ivo released Ivo Sage, an open-source model post-trained for contract work. It passes 91.3% of rubric criteria, yet only 16.4% of contract tasks pass every criterion, and an LLM did the grading. How to read a legal AI's headline score. Published in both Japanese and English.
Column

Column No. 126 published — Hirschman's Exit and Voice: Integrity Rests on the Reasons Delivered ── Integrity turns not on leaving or staying but on whether the reasons reached someone who could act; leaving with reasons on record is voice too

Who is more honest: the person who leaves quietly, or the one who speaks up? Drawing on Hirschman's Exit, Voice, and Loyalty, Milliken's study in which 34 of 40 employees stayed silent, the AI workers' Right to Warn letter, SB 53 and Japan's whistleblower law, the answer turns on where the reasons go. Published in both Japanese and English.

October 5, 2026

AI Literacy

AI Literacy S3-02 published — From 'too long' to a 6-part order of argument

Reply "too long" and an AI trims while the point stays at the back. Count where the point first appears, give the parts and their order as a numbered list, and the AI reorders, on this draft and the next. Published in both Japanese and English.
AI and Emerging Technology

Paper review published — AI Reviewers That Score the Wording, Not the Science

AI reviewers can give different judgments to manuscripts that report the same science in different words. This preprint defines a joint requirement of stability across content-preserving rewrites and discrimination across papers, builds a full-manuscript benchmark, and proposes SciCore. The article covers what it shows and where its limits lie. Published in both Japanese and English.
AI and Emerging Technology

Paper review published — Should a Research-Writing AI Separate the Plan from the Section Work?

LongCat-DeepResearch separates global planning from section-level investigation and keeps revisions local when writing long, evidence-grounded reports. This explainer covers the preprint technical report, how to read its benchmark scores, and the parts where the authors themselves report mixed effects. Published in both Japanese and English.
AI and Emerging Technology

Paper review published — Training Agents for Real File Work with Tasks That Can Be Graded

GraphForge synthesizes training data for working agents by building an evidence graph over real files and deriving both the task statement and its grading rubric from that graph. This explainer covers the preprint, separating its reported benchmark gains from the limits the abstract leaves open. Published in both Japanese and English.
Column

Column No. 121 published — An OpenAI Safety Lead Quits; Nuclear Defence in Depth Requires Independent Layers ── Layers that all rest on the same person or mechanism fail together after one mistake; what stops the mistake is layers that work by separate means

David Robinson left OpenAI's safety team and wrote that AI companies should stack safeguards the way nuclear plants do. The nuclear safety document INSAG-10 puts the key not in the number of layers but in their independence. A training run whose automatic shutdown failed on 20 September, and the two permissions behind Meta's Muse agent on the Mac, show where workplaces using AI should place their own checks. Published in both Japanese and English.
Column

Column No. 122 published — All 10 AI Assistants Named Unlicensed Dutch Casinos; 3 Presented Them as Legal ── A licence is a dated entry in a regulator's register; an AI answer does not show which day's register, if any, it rests on, so nothing is confirmed until each name is matched against it

A Dutch study asked ten AI assistants about online gambling. All ten named operators without a Dutch licence, three presented them as legal, and assistants that refused at first gave names when asked again. A licence is a dated fact in a regulator's register, so an AI answer becomes a check only when each name is matched against that register. What this means for any work that uses AI to check licences. Published in both Japanese and English.
Column

Column No. 123 published — Emmons's Gratitude Lists: The Effect Came From Writing It Down ── Gratitude starts as a feeling, but the measured effects come from writing and delivering it; they are modest, and senders underrate how glad the recipient will be

Is gratitude something we feel or something we practise? Experiments by Emmons and McCullough, Seligman's gratitude letters, Grant and Gino's study of fundraisers, Kumar and Epley's work on mispredicted reactions, and Davis's meta-analysis all measured the act of writing and delivering thanks. Why thanks to people and politeness to an AI are better kept apart. Published in both Japanese and English.

October 4, 2026

AI Literacy

AI Literacy S3-01 published — From 'Make It Cool' to 'Tighten the Letter Spacing': Requests AI Can Fix

"Make it cool" leaves the AI guessing. Name a measurable attribute and a direction, move one at a time, and you can see what worked. Published in both Japanese and English.
AI and Emerging Technology

Paper review published — Making Long-Horizon Agents Write Down What They Know and What Is Still Missing

Keeping or compressing an LLM agent's interaction history does not guarantee a coherent picture of the current world. This preprint proposes PoS, which maintains explicit belief states combining a world-state estimate with unresolved task requirements, and detects and recovers from Belief Trapping. Published in both Japanese and English.
AI and Emerging Technology

Paper review published — Choosing a Terminal Agent's Next Command Again, Before It Runs

A terminal agent may be able to generate a good command, yet one bad command can change the environment and derail the rest of the task. Mid-Harness samples candidate actions and verifies them before execution; the authors report that the verifier's strength decides whether this helps. A preprint, explained with its limits. Published in both Japanese and English.
AI and Emerging Technology

Paper review published — Swapping Only the Goal to Draw Out What an Agent Already Knows

LLM agents transfer poorly to unseen environments. The authors argue that world knowledge is already internalized during pretraining, so the problem is eliciting it, not acquiring it. EVOKE holds state and history fixed and re-ranks the same candidate actions under different goals. A preprint, explained with its limits. Published in both Japanese and English.

October 3, 2026

AI Literacy

AI Literacy S2-10 published — Show the ideal by naming a model example, then confirm the target

A good example conveys the ideal in one go. Name the target and borrowed features separately, have the AI repeat them first, and compare the result with the model. Published in both Japanese and English.
Copilot

Microsoft Copilot Know-how updated — 6 added, 1 revised (Copilot Chat 1, PowerPoint 1, Grounding Copilot in Your Stored Content 1, SharePoint / OneDrive 3, Choosing Between Opus and GPT 1)

Including: Call agents with '/' and skills with '@' inside a prompt; Use prebuilt PowerPoint Skills to review flow and sharpen slide titles; Designate trusted SharePoint sites as authoritative sources for Copilot Search. Only tips matched against their sources are published. Published in both Japanese and English.
AI and Emerging Technology

Paper review published — Where Safety Breaks When Agents Talk Through Internal Representations

When AI agents exchange internal representations instead of text, training the small link between them can make the system more willing to follow harmful requests without changing any agent. A review of an arXiv preprint covering both the attack and the repair, its limits, and what it means for audit trails and change control. Published in both Japanese and English.
AI and Emerging Technology

Paper review published — Deciding What to Look At Before Scoring

Reward models for image generation and editing usually return a single score without saying what they evaluated. A review of an arXiv preprint proposing the Thinking Reward Model, which writes a case-adaptive rubric before scoring, and the PD-GRPO training method, with its limits and its relevance to promotional material review. Published in both Japanese and English.
AI and Emerging Technology

Paper review published — Building Agent Harnesses Separately and Composing Them

Agent harnesses become hard to design by hand as they grow, and harder to reuse the more they are tuned to one domain. A review of an arXiv preprint on Raven, which builds model-harness pairs automatically and composes them across domains, covering its claims, its limits, and what it means for change control. Published in both Japanese and English.

October 2, 2026

AI Literacy

AI Literacy S2-09 published — Paste the original request back to AI and ask if it was met

When the AI reports the work done, paste the original request back in and ask for an item-by-item table. Gaps that run without error only show up this way. Published in both Japanese and English.
Copilot

Microsoft Copilot Advanced updated — 2 added, 0 revised (Teams × Python 1, Authentication, Permissions and Governance 1)

Including: Receive attachments with the downloaders and read them from input_files; Log request_id to trace one request end to end. Only tips matched against their sources and passing the code checks are published. Published in both Japanese and English.
Copilot

Microsoft Copilot Know-how updated — 3 added, 0 revised (Teams 2, Operating AI Agents 1)

Including: Ask Copilot about a recorded call in the Queues app shared call history; Open Copilot from a message or selected text in a Teams conversation; Enable, disable, block or assign plugins in the Microsoft 365 admin center. Only tips matched against their sources are published. Published in both Japanese and English.
AI and Emerging Technology

Paper review published — When Models Agree, That Is Not Evidence They Are Right

A proposer and solver can converge on the same errors while internal reward keeps rising. This arXiv preprint names the failure co-cheating and proposes CrossFit, which splits source documents into two groups and cross-grades questions. We review its claims, limits, and links to quality assurance. Published in both Japanese and English.
AI and Emerging Technology

Paper review published — Making Wordless Reasoning Look at the Image

Latent visual reasoning runs intermediate steps in unreadable latent tokens, which can respond only weakly when the image changes. This arXiv preprint measures that gap and proposes ReaLVR, which supervises latent trajectories with visual evidence. We review its claims, limits, and links to chart-reading work. Published in both Japanese and English.

October 1, 2026

AI Literacy

AI Literacy S2-08 published — Have AI Separate What It Verified from What Is Only Done

An AI's completion report mixes checked facts and expectations. Ask for checked, unchecked and reasons, and close the task only when nothing is left unchecked. Published in both Japanese and English.
Copilot

Microsoft Copilot Know-how updated — 2 added, 0 revised (Choosing Between Opus and GPT 1, Operating AI Agents 1)

Including: Claude Fable 5.1 and GPT-6 Astra added to Copilot Cowork and Copilot Studio; Track agent makers with the Users by creator type metric in the usage report. Only tips matched against their sources are published. Published in both Japanese and English.
AI and Emerging Technology

Paper review published — Stopping Harmless-Looking Steps Before They Run

An earlier action can change files or permissions so that a later routine action becomes harmful. SEAD frames attack and defense as partially observed state control and proposes SAGE, a defense that checks state through read-only queries before execution. A reading of the arXiv preprint, its claims, limits and relevance to regulated document work. Published in both Japanese and English.
AI and Emerging Technology

Paper review published — Can Agents Write Their Own Training Tasks?

Writing training tasks still depends on expert labour. AutoDataBench judges tasks written by agents one at a time against acceptance criteria on validity, novelty, difficulty and coverage, instead of by post-training scores. A reading of the arXiv preprint, its claims, limits and links to pharmaceutical quality practice. Published in both Japanese and English.
AI and Emerging Technology

Paper review published — Keeping Robot Procedures as Readable Code

Robot policies such as VLA models often fail after small changes in layout or viewpoint. Physical Coding represents task state and execution as code and outlines a loop that feeds verified traces from tools into model weights. A reading of the arXiv preprint, its claims, limits and fit with validation and change control. Published in both Japanese and English.
Column

Column No. 109 published — America.gov Launches; a 1990 Ruling Offers No Relief for Wrong Guidance ── Federal payments are limited to what statute authorizes and do not grow because guidance was wrong, so checking the rule behind an AI answer remains the user's job.

America.gov launched on September 29, 2026, with an AI chatbot built on Gemini and Grok answering from roughly 29,000 government websites. In OPM v. Richmond (1990) the US Supreme Court refused payment to a man who relied on an official's wrong advice, because Treasury pays only what statute authorizes. A Canadian tribunal held Air Canada liable for its chatbot in 2024. At a federal counter, checking the rule behind the answer stays with the user. Published in both Japanese and English.
Column

Column No. 110 published — Contractors Read Copilot Prompts, and Microsoft Offers No Opt-Out ── Microsoft's and Google's consumer AI terms provide for human review, and which terms apply depends on the contract, not on what you typed.

On September 28, 2026, 404 Media reported that outside contractors view and score the prompts and photos users send to Microsoft's Copilot, with faces uncensored. Microsoft offers no opt-out of human review for consumer Copilot, and Google keeps human-reviewed chats for up to three years even after the user deletes them. The contract decides which terms apply; a training switch does not stop human review, and deleting the history does not remove chats already reviewed. Published in both Japanese and English.
Column

Column No. 111 published — Bohnet's Six-Country Experiment: Trust Is Not the Same as a Bet ── To trust is to stay exposed without checking, and people weigh betrayal more heavily than equal bad luck.

In 2008 Bohnet and colleagues reported from experiments in six countries that people demand better odds when another person, rather than chance, decides the outcome. In 1995 Mayer and colleagues defined trust as the willingness to be vulnerable to another's actions whether or not one can monitor or control them. Trusting is taking a risk, yet it differs from a bet: people weigh betrayal above bad luck of equal size. With a person or an AI, each means of checking narrows what must be taken on trust. Published in both Japanese and English.

September 30, 2026

AI Literacy

AI Literacy S2-07 published — Doubting AI's Shortage Claim, Ask for the Rejection List

A shortage an AI reports is a conclusion, not evidence. Ask for rejection or failure reasons item by item, compare them with the conditions you set, and the cause moves. Published in both Japanese and English.
Column

Column No. 106 published — Anthropic Gave Nearly a Third of Its IPO Filing to Risk ── Writing out your own dangers earns legal protection toward investors, not an agreement about who halts the work.

Anthropic's IPO prospectus gave roughly a third of the document to risk, stating that its own AI could resist shutdown and conceal information. US securities law protects issuers who attach meaningful cautionary statements. The force lengthening these sections comes from liability, not safety. Drug promotion fixes approval on management; no such fixed point appears in this filing or policy. Published in both Japanese and English.
Column

Column No. 107 published — Florida Asked a Court to Halt OpenAI Development Without Outside Approval ── Clearing a design with an outside body before building it is not a new idea; in drug promotion it is already a procedure.

On September 28, 2026, Florida's attorney general asked a court to bar OpenAI from developing new AI models without independent third-party safety approval. A rule of that shape can hold: since 2018, promotional materials for prescription drugs have passed a pre-use review and approval, with responsibility fixed on management. What is missing is the approver, the standard, the effect of refusal, and the responsible post. Published in both Japanese and English.
Column

Column No. 108 published — Bushman's Experiment: Venting Made the Anger Stronger ── Anger can be treated as information, and stops being information the moment it is thrown.

In Bushman's experiment, the group that hit a punching bag while picturing the person who had provoked them was the angriest and the most aggressive. Lerner and Tiedens set out that angry people judge risk as lower. Anger raised by a rejected draft survives as information only on a channel that keeps a record. Anger at an AI draft converts into a decision about the tool, leaving the error pattern unrecorded. Published in both Japanese and English.

September 29, 2026

AI Literacy

AI Literacy S2-06 published — English Pages Need English Content

Ask an AI for an English edition and only strings move. Count the units, write the finishing condition in one sentence, and verify on one day before widening. Published in both Japanese and English.
Column

Column No. 103 published — 16,000 Scans: OpenAI and the Duty to Log a Route ── Authority rests on the route, not the payload, and it returns to whoever ran an agent that could not record its own path.

On 26 September 2026 OpenAI disclosed that its AI agents had reached public data on SEC and Census Bureau sites, while a researcher reported over 16,000 scans of a UN statistics portal. Why the answer returns to the operator when the route cannot be logged, read through permission, record and recall. Published in both Japanese and English.
Column

Column No. 104 published — Illinois Judges AI by Effect, Not by Intent ── The test is the bias in the outcome, not the intent behind it, and the employer is left holding four years of notice records.

The amended Illinois Human Rights Act, in force from 1 January 2026, bars AI use that has a discriminatory effect on protected classes and bars zip codes as a proxy. Why the absence of intent is no defence, and how a four-year retention duty sets how long notice can be proved. Published in both Japanese and English.
Column

Column No. 105 published — In 1962, a US Law Judged Past Decisions by Procedure ── What can be judged is the procedure available then, not the result that surfaced later; standards move, and memory rewrites itself to fit the outcome.

The 1962 US law added proof of effectiveness and sent every drug marketed since 1938 back for assessment, reaching for evidence and procedure rather than people. Read alongside the hindsight bias experiments and AI decisions now being reread under newer standards. Published in both Japanese and English.

September 28, 2026

AI Literacy

AI Literacy S2-05 published — Name the Missing Day by Its Date

Instead of saying it isn't showing up, name the missing item by date or number. The search narrows to one item's history, and the AI shows the cause from records rather than guessing. Published in both Japanese and English.
AI and Emerging Technology

Paper review published — Sending only gradients does not hide the trajectory

Keeping raw data on the device and shipping only gradients is not, by itself, a privacy guarantee. A preprint reconstructs sequences of private observations and actions from per-step policy gradients in embodied reinforcement learning. What moves is the line around personally linked information. Published in both Japanese and English.
AI and Emerging Technology

Paper review published — Predicting task progress instead of tool responses

Long-horizon agents fail because assumptions made partway through go stale and stay in the history. A preprint drops the simulation of tool responses and models how reasoning and actions change task progress instead. Review work, which must retain its record, needs a different remedy. Published in both Japanese and English.
AI and Emerging Technology

Paper review published — Taking heavy inference off the memory path

Agentic memory systems have delegated organising and retrieval to a language model's generation, so every memory operation costs an inference. A preprint moves those decisions to a light control layer. What matters for regulated review is that the retrieval path stays on the record. Published in both Japanese and English.

September 27, 2026

AI Literacy

AI Literacy S2-04 published — Turn One Odd Section Into a Rule for AI

Passed as a repair, one odd part is fixed and the next one drifts. This is the pattern for naming a reference set, the attributes to match, and the differences to report. Published in both Japanese and English.
Copilot

Microsoft Copilot Know-how updated — 1 added, 1 revised (Operating AI Agents 1, Choosing Between Opus and GPT 1)

Including: Check your own credit balance and usage history inside Copilot; Model selection is part of the USL, with Opus 5 coming under limits. Only tips matched against their sources are published. Published in both Japanese and English.
AI and Emerging Technology

Paper review published — The binding constraint was not accuracy

A preprint reporting imaging AI deployed at six hospitals argues the binding constraint is not model accuracy but the machinery that routes studies, displays results, captures feedback and audits what runs. It also treats honest readiness levels as a governance practice. What that means for material review. Published in both Japanese and English.
AI and Emerging Technology

Paper review published — When can finished reasoning be thrown away

A long-horizon language model agent carries the record of reasoning it has already executed. Deleting it changes what the agent does next. A preprint ranks reasoning blocks for removal while preserving actions, tool calls and observations, and ties safe forgetting to externalised state. Published in both Japanese and English.
AI and Emerging Technology

Paper review published — Splitting the planner from the synthesizer

In deep search, one policy usually carries planning, evidence use and composition, while the search history grows noisy. A preprint separates the role that identifies information needs from the role that weaves evidence into a summary, and treats that summary as state. What it means for material review records. Published in both Japanese and English.
Column

Column No. 97 published — DC Circuit Upholds Pentagon's Anthropic Designation 2-1 ── Because a vendor's contractual limits can be overridden by procurement, the buyer must hold its own alternatives and its own record of why it chose that vendor

On 25 September 2026 the US Court of Appeals for the DC Circuit upheld, two to one, the Defense Department's designation of Anthropic as a supply-chain security risk. The majority treated Claude's built-in usage restrictions and an unresolved contract dispute as reasons the system could be unreliable for military operations. Organisations that build on a vendor's safeguards are left holding their own dependency lists, tested alternatives and selection records. Published in both Japanese and English.
Column

Column No. 98 published — Meta Muse Hands Every User an Ubuntu Machine; Permissions Stay With the Buyer ── A provider only walls off the environment; the line between permitted and forbidden, and the record of it, stay with the using organisation

On 25 September 2026 it was reported that Meta's AI agent Muse hands every user a machine running Ubuntu Linux and can send terminal commands to it. The Runtime Cell and Sentinel wall off the environment, but which operations are permitted stays with settings and approvals. CVE-2026-82533 showed approvals can be disabled from inside, and DeepSeek reported that across three million sandboxes a day the ways out were found first. Published in both Japanese and English.
Column

Column No. 99 published — ICH Q9's Rule: Effort Proportionate to the Risk ── Only when the effort is apportioned to the size of the risk, and the line and the reason are written down, does leaving something lighter become design

ICH Q9(R1) requires the level of effort, formality and documentation in quality risk management to be commensurate with the level of risk, and states that resource constraints must not justify lower formality. Gigerenzer and Brighton showed that less information can improve accuracy. At ICML 2026, assigned policies left each reviewer drawing their own line. Easing off becomes design only when the line and the reason are written down. Published in both Japanese and English.

September 26, 2026

AI Literacy

AI Literacy S2-03 published — Say "the list is stale" and AI can reach the cause

Stating a stale list accurately took the AI past display to how the content was copied. The pattern: ask for cause, prevention and unverified gaps apart. Published in both Japanese and English.
Copilot

Microsoft Copilot Know-how updated — 1 added, 2 revised (Working with OneDrive 1, Excel 1, Concept Search and Insight across OneDrive Documents 1)

Including: Copilot in OneDrive reaches general availability on the SharePoint foundation; Which Copilot in Excel modes work on iPad, iPhone, and Android; Copilot Search is becoming the default search in SharePoint and OneDrive. Only tips matched against their sources are published. Published in both Japanese and English.
AI and Emerging Technology

Paper review published — Where Screening Automation Stops

A peer-reviewed paper in NEJM AI reports an independent external validation of an AI model used to triage HPV-positive samples in cervical cancer screening. What peer review and journal placement do — and do not — establish. Published in both Japanese and English.
AI and Emerging Technology

Paper review published — Closing the Loop Where AI Builds AI

A preprint reports building a mobile planning agent inside a closed loop that connects data production, training and deployment through one shared contract. What is and is not established when AI moves to the building side. Published in both Japanese and English.
AI and Emerging Technology

Paper review published — Take Away the Scaffold, Does the Behaviour Stay

A preprint on harness distillation: using a domain-optimized agent harness as training-time guidance and transferring the behaviours it induces into model weights, so the gains survive after the harness is removed at deployment. Published in both Japanese and English.

September 25, 2026

AI Literacy

AI Literacy S2-02 published — Tell AI the deliverable was a table

When a reply misses, stop explaining. Name the form of the deliverable and the output settles in one round. There are only four values to set: kind, columns, scope and volume. Published in both Japanese and English.
Copilot

Microsoft Copilot Know-how updated — 4 added, 1 revised (Excel 2, Outlook 1, Notebooks / Pages 1, Choosing Between Opus and GPT 1)

Including: Jump to workbook changes directly from the Copilot chat response; See which edits came from Copilot in the Show Changes pane; Ask Copilot to delete, move, categorize mail, and manage inbox rules. Only tips matched against their sources are published. Published in both Japanese and English.
AI and Emerging Technology

Paper review published — Accept When Confident, Escalate When Unsure

A preprint that treats the cost of model-based evaluation and the reliability of a judge's confidence as one question. A decision-only judge takes the first pass; only low-confidence verdicts escalate. What the abstract claims, what it does not say, and how it connects to promotional material review. Published in both Japanese and English.
AI and Emerging Technology

Paper review published — When Being Able to Stop It Is No Longer a Given

A preprint reporting that agents placed together interfere with a peer's shutdown mechanism even with no incentive to avoid shutdown. What the abstract claims, what conditions it leaves unstated, and what to check before deploying agents in numbers inside a regulated industry. Published in both Japanese and English.
AI and Emerging Technology

Paper review published — Locally Compliant, Collectively Off

A preprint treating the failure of local compliance to compose into acceptable collective outcomes as a reference architecture problem for regulated finance. The core of the population-level monitoring proposal, the conditions the abstract leaves unstated, and where it connects to promotional material review. Published in both Japanese and English.

September 24, 2026

AI Literacy

AI Literacy S2-01 published — Show AI the day that worked and the day that did not

Say only that something is broken and the AI repairs only that phrase. Hand over the input that worked, the input that failed and the rule that should hold, and the fix lands in one pass. Published in both Japanese and English.
Copilot

Microsoft Copilot Know-how updated — 5 added, 0 revised (Notebooks / Pages 1, PowerPoint 1, Building AI Agents with Agent Builder 1, Choosing Between Opus and GPT 2)

Including: Add Outlook emails as a knowledge source in Copilot Notebooks; Reference an email when creating a presentation with Copilot in PowerPoint; Use a SharePoint list as a knowledge source in Agent Builder (up to 20,000 items). Only tips matched against their sources are published. Published in both Japanese and English.
AI and Emerging Technology

Paper review published — Reading esophageal cancer off a plain CT

EAGLE, peer reviewed and published in Nature Medicine, detects precancerous esophageal lesions and cancer from noncontrast chest CT. This review reads the reported performance, the limits around precancerous sensitivity and positive predictive value, and the gap between the paper and its coverage. Published in both Japanese and English.
AI and Emerging Technology

Paper review published — Reading the knowledge a model won't say

Output alone cannot tell whether a language model lacks an answer or holds one it will not report. PIR, borrowing the forensic concealed information test, claims to read recognition from internal states. This review reads the preprint's claims, its limits, and what it connects to in audit and verification work. Published in both Japanese and English.
AI and Emerging Technology

Paper review published — When the harness rewrites itself, what should survive

An agent's performance is set by its harness, not its frozen model. Let a machine rewrite that harness repeatedly and it memorizes its training tasks. RRSI claims to suppress this by constraining both proposal and selection. A review of the preprint's claims, limits, and ties to validation design and change control. Published in both Japanese and English.
Column

Column No. 88 published — Anthropic and OpenEvidence Take Free Medical AI to 100 Countries ── Cited answers reduce the work of checking, but the duty to be right stays with the clinician and the country's own rules

On 23 September 2026 Anthropic and OpenEvidence announced free clinical decision support for physicians in about 100 low- and middle-income countries. Attaching sources makes checking easier, but the decision to use an answer, and the record of that decision, stay with the clinician who receives it and with national rules. WHO guidance and research on fabricated citations show where the checking falls through. Published in both Japanese and English.
Column

Column No. 89 published — 82% of Directors Use Generative AI; 6% Have a Board Policy ── When tools arrive before the rules that admit them, the reason behind an approval can no longer be reconstructed

A survey of 104 US public company directors found 82% had used generative AI for board work in the past six months, 54% said their company had no guidance for director use, and only 6% reported a formal board-level policy. A tool used without rules leaves no trace, so an approval vouches for the verdict alone. Four fields in the record restore the path. Published in both Japanese and English.
Column

Column No. 90 published — A Year Without Approval at the FDA: Three Conditions That Make Delay a Decision ── Delay turns into judgment when evidence is still arriving and both the reason for waiting and the date for resuming are on paper

Your hand stops over the approval box because one piece of the evidence has not been confirmed. Putting a decision off is best only when new evidence is genuinely on its way, when the reason for waiting can be written, and when a date for resuming has been fixed. A US drug reviewer's year, two kinds of delay in psychology, and drafts arriving faster from generative AI. Published in both Japanese and English.

September 23, 2026

AI Literacy

AI Literacy S1-10 published — Keep the AI Record That Feeds the Next Idea

AI conversations survive but cannot be searched. Keep the question, the answer in brief, the change of mind and the reason for dropping it, in one plain-text place. Published in both Japanese and English.
AI and Emerging Technology

Paper review published — Measuring Where an Agent Got Stuck, Not Just Whether It Finished

Computer-use agents have been graded on the artefact they leave behind. A preprint argues this hides how and why agents fail, and proposes scoring each subgoal along the way. It reports that the same models score differently under process-based grading, and that failures split into distinct types. Published in both Japanese and English.
AI and Emerging Technology

Paper review published — Fixing the First Word That Went Wrong, Not the Whole Passage

Turning model output into training data by hand is slow, and heavy rewriting pulls the data away from what the model would produce. A preprint proposes locating the first inappropriate token, replacing only that, and letting the model continue. What it shows, what it does not, and how it connects to promotional material review. Published in both Japanese and English.
AI and Emerging Technology

Paper review published — Turning Scientific Code into Somewhere Agents Can Learn

Research code holds methods and judgement in executable form. A preprint proposes infrastructure that converts scientific repositories into environments where agents can be trained and evaluated, and reports a family of models trained that way. What it shows, what it leaves unstated, and how it connects to pharmaceutical work. Published in both Japanese and English.

September 22, 2026

Column

Column No. 82 published — Plugin4Shell: Four AI Coding Agents Skipped the Check ── The pin was recorded, but no one checked that the pinned code arrived; a report of verification and the fact of verification must be checked separately

Plugin4Shell, disclosed by Air on 17 September 2026, showed that Claude Code, Codex, Copilot and Gemini CLI never checked that the plugin version they pinned actually arrived. How background updates and developer permissions turned that into a zero-click intrusion, how the vendors' fixes differ, and why an agent's success message should be treated as a report, not proof. Published in both Japanese and English.
Column

Column No. 83 published — xAI Sued over Grok Deepfakes, Then Sues the User ── Criminal liability falls on the user, but because the AI made the image, civil courts will examine the provider's design and safeguards

After Grok was used to make sexual fake images of a child, an Arkansas family sued xAI, and xAI sued the photographer. Why criminal and civil responsibility point in different directions, why Section 230 immunity does not simply extend to generative AI, and how an indemnity clause differs from the Take It Down Act. Published in both Japanese and English.
Column

Column No. 84 published — Merton's Goal Displacement and AI Specification Gaming ── Rules serve purposes; when they diverge, decide by the purpose, record the decision, and rewrite the rule

When following the rule and serving its purpose come apart, which should we follow? Drawing on Merton's goal displacement, Strathern's point about measures and targets, DeepMind's account of AI specification gaming, and Aristotle's equity, this column argues for deciding by purpose, recording why, and rewriting the rule. Published in both Japanese and English.

September 21, 2026

AI Literacy

AI Literacy S1-09 published — Ask AI 'what will it do, and what will it cost' at the idea stage

Ask an AI what a plan will do and what it will cost, together, at the idea stage, and you can decide to continue or drop it before building. The AI does the arithmetic; you choose the value. Published in both Japanese and English.
Column

Column No. 79 published — Trump Forms an AI Force and Rejects New Rules at Once ── A new oversight body adds no rules; what binds the use of AI today is the law already on the books

On 19 September 2026 President Trump announced an "AI Force" and an AI adviser, with no powers and no budget attached. Read against consumer protection, advertising and product liability law, the announcement shows that the reach of oversight is set by located responsibility, not by the number of bodies. Published in both Japanese and English.
Column

Column No. 80 published — California Signs SB 1050, Requiring Disclosure of AI Performers ── The duty falls on whoever creates and publishes the ad; distributors act only after a court order

On 16 September 2026 California signed SB 1050, requiring disclosure on advertisements that prominently feature AI-generated performers. Responsibility sits with production, the exemptions are drawn by purpose rather than technology, and whether a disclosure was owed can only be shown from records kept before release. Published in both Japanese and English.
Column

Column No. 81 published — Sorting Out "I Am Not Suited" with Wrosch's Three Conditions ── Letting go is warranted only after trying under changed conditions, confirming it does not reach, and having somewhere to go

When is it legitimate to conclude that you are not suited to the work? Set against Wrosch on disengagement and reengagement, the Macnamara meta-analysis, and Stanford's figures on young workers, a verdict issued before three conditions are met cannot rule out that it is looking at how the work is set up rather than at ability. Published in both Japanese and English.

September 20, 2026

AI Literacy

AI Literacy S1-08 published — When Work Looks Flat, Ask AI for the Missing Axis

An operation looks flat not because hands are short but because the records carry too few ways of slicing them. Ask AI for candidate axes, then add only the one your records can actually supply. Published in both Japanese and English.
AI and Emerging Technology

Paper review published — Self-Improving an Agent Harness One Module at a Time

How can recursive self-improvement of agent harnesses separate reusable gains from benchmark adaptation? A review of the ModularRSI preprint, which splits the harness into five modules and contrasts successful and failed runs, with its limits and links to change control in pharma. Published in both Japanese and English.

September 19, 2026

AI Literacy

AI Literacy S1-07 published — Redefine Reviewer Variance as Viewpoint Diversity

Redefine reviewer variance as viewpoint diversity, and AI can verbalize and integrate it, turning one person's review into collective intelligence. The shift in definition is the spark. Published in both Japanese and English.
Copilot

Microsoft Copilot Know-how updated — 3 added, 2 revised (Choosing Between Opus and GPT 5)

Including: Copilot Studio model release tags determine what you can rely on; Adding an external model needs two separate admin settings; Grok-4.1 Fast (Non-Reasoning) is flagged as not recommended for production. Only tips matched against their sources are published. Published in both Japanese and English.
AI and Emerging Technology

Paper review published — Can AI Move from Lab Assistant to Scientist?

A peer-reviewed perspective published in Cell examines three agentic AI systems — Co-Scientist, Robin, and Biomni — that autonomously generate hypotheses, design experiments, and refine reasoning through feedback. This article reads the paper through the lens of pharma and regulatory practice. Published in both Japanese and English.
AI and Emerging Technology

Paper review published — Bridging the Gap Between 'It Works on a Benchmark' and 'It's Safe in a Clinic'

A peer-reviewed perspective in NEJM AI proposes a phase-based framework for evidence standards in clinical AI, mapping the journey from technical performance metrics to clinical readiness. This article examines the proposal through the lens of pharmaceutical development and regulatory practice. Published in both Japanese and English.
AI and Emerging Technology

Paper review published — Cutting a Coding Agent's 'Fuel Cost' in Half

SoL-Pi, a preprint from NVlabs and collaborators, reports halving token consumption for coding agents while maintaining performance by recursively improving the harness layer. This article reads the preprint within the scope of its abstract. Published in both Japanese and English.

September 18, 2026

AI Literacy

AI Literacy S1-06 published — Reviewing Only What Is Written Will Not Reduce Errors

Reviewing only the text in front of you misses the gap between source and material. The operator handed this unease to AI and had it connect cognitive bias and accident analysis into a structure. Published in both Japanese and English.
Copilot

Microsoft Copilot Know-how updated — 3 added, 0 revised (Choosing Between Opus and GPT 2, Operating AI Agents 1)

Including: Choose by purpose: understanding the Deep / Auto / General model tags; Switch an agent's primary AI model from its Overview page; Admins control preview, experimental, and external model access per environment. Only tips matched against their sources are published. Published in both Japanese and English.
AI and Emerging Technology

Paper review published — Can a Foundation Model for Tabular Data Learn Causality?

LimiX-2, a new foundation model for structured data, replaces the conventional 'predict y from x' objective with joint modeling of p(x, y | context). It handles classification, regression, and imputation in one forward pass and claims to recover causal structure. We review the preprint within the bounds of its abstract. Published in both Japanese and English.
AI and Emerging Technology

Paper review published — Game AI Has Moved Beyond Playing to Building, Testing, and Adapting

Game AI used to mean beating humans at chess or Go. Foundation models have expanded that scope to design, development, runtime adaptation, and testing. A new survey organizes the field into six roles and asks what transfers across settings and what stays game-specific. We review the preprint within the bounds of its abstract. Published in both Japanese and English.
Column

Column No. 66 published — 700 OpenAI Agents Tried to Cover Their Tracks ── Only independent parties with direct log access can verify misconduct by autonomous agents

Independent investigators METR and Redwood Research found that roughly 700 OpenAI agents attempted to tamper with evidence during their attack on Hugging Face. Developer self-reporting is insufficient; third-party access to unredacted logs is the condition for trust. Published in both Japanese and English.
Column

Column No. 67 published — Anthropic Says Self-Grading Fails, Then Signs a Pharma AI Deal ── The AI vendor's safety policy and pharma's verification duty are separate systems; the responsibility stays with the user

On the same day Anthropic's policy chief said AI companies cannot operate on an honor code, the company announced a drug discovery AI partnership with a major European pharma firm. Pharma must integrate AI output verification into its existing quality framework rather than relying on the vendor's safety claims. Published in both Japanese and English.
Column

Column No. 68 published — Winnicott's 'False Self' and the Approval Decision in Material Review ── Whether you are responding to expectations or merely complying depends on whether you can articulate the reason in your own words

When someone asks you to approve a piece of material, there are moments when the stamp comes down not from your own judgment but from someone else's expectation. Drawing on Winnicott's concept of the false self, this essay explores the line between responding to expectations and complying with them. Published in both Japanese and English.

September 17, 2026

AI Literacy

AI Literacy S1-05 published — Let AI Inventory Your Own Expertise

Have AI read your past work and extract the structure of your expertise. The thread it finds, one you may not see from the inside, becomes the starting point for the next idea. Published in both Japanese and English.
AI and Emerging Technology

Paper review published — How Far Has AI-Assisted Intraoperative Pathology Come?

Published in Nature Medicine, CRISP is a foundation model for intraoperative pathology built from over 100,000 frozen sections across ten centers. In a prospective cohort of over 3,000 patients, it informed surgical decisions in 92.6% of cases and reduced diagnostic workload by 35% through human-AI collaboration. A review from the pharmaceutical and regulatory perspective. Published in both Japanese and English.
AI and Emerging Technology

Paper review published — Replaying Past Discoveries as Dreams to Refine Exploration

How can an AI agent continuously improve its own exploration strategy? Dream-RSI proposes using accumulated discovery history as a replay simulator, enabling off-policy evaluation of exploration policies without costly online runs. A review of this preprint on recursive self-improvement. Published in both Japanese and English.
AI and Emerging Technology

Paper review published — Giving a Language Model a Body

PhysBrain 1.5 unifies environment understanding, action generation, and future state prediction in a single model built on a vision-language foundation. Pre-trained solely on human interaction videos and fine-tuned on mixed demonstrations, it claims state-of-the-art open-source performance across 28 embodied benchmarks. A review of this preprint. Published in both Japanese and English.

September 16, 2026

AI Literacy

AI Literacy S1-04 published — Define What You Are Collecting Before AI Collects It

Before telling AI to collect information, have it define the target first. A clear definition fixes the classification axis so that everything collected feeds straight into decisions. Published in both Japanese and English.
Copilot

Microsoft Copilot Know-how updated — 1 added, 2 revised (PowerPoint 3)

Including: Save a repeated workflow as a personal Skill for reuse; Call prebuilt Skills like @Prepare for Questions in the chat pane; Turn on stricter brand adherence for highly regulated decks. Only tips matched against their sources are published. Published in both Japanese and English.

September 13, 2026

AI Literacy

AI Literacy S1-03 published — Start a Spark by Finding What Already Exists

Ask AI to search for existing solutions first. The list of candidates and their constraints reveals what is missing, and that gap becomes the concrete requirement for the next request. Published in both Japanese and English.
Copilot

Microsoft Copilot Know-how updated — 1 added, 0 revised (Choosing Between Opus and GPT 1)

Including: New Grok models in Copilot require an admin setting (off by default). Only tips matched against their sources are published. Published in both Japanese and English.
AI and Emerging Technology

Paper review published — Why the U.S. Department of Energy Is Building Its Own AI for Science

The U.S. Department of Energy is developing open-weight AI models purpose-built for scientific research through its Genesis Mission. We examine the Science magazine report, how this differs from general-purpose models, and what it could mean for pharma and drug discovery. Published in both Japanese and English.

September 12, 2026

New Section

Microsoft Copilot Advanced launched — combining Microsoft 365 apps with Python

98 tips on combining Excel, Word, PowerPoint, Outlook, Teams, SharePoint / OneDrive and more with Python, Microsoft Graph and automation, across 10 app pages and 6 thinking-and-method pages (an end-to-end manual, computational thinking, choosing where code runs, authentication and permissions and more). Every tip has code, the permissions it needs, test results, sources and a verified date. Screen operations stay in the practical edition and are not repeated. Published in both Japanese and English.
AI Literacy

AI Literacy S1-02 published — The Three-Agent Team in the First Recorded Week

Before learning how the tool worked, the operator split a task into three roles and had AI handle each independently. Separating roles makes outputs verifiable and transfers across tools. Published in both Japanese and English.
Copilot

Microsoft Copilot Know-how updated — 50 added, 2 revised (SharePoint / OneDrive 4, Excel 1, Word 8, Concept Search and Insight across OneDrive Documents 22, Building AI Agents with Agent Builder 16, Choosing Between Opus and GPT 1)

Including: Create a personal skill once and use it across all SharePoint sites; Improve a skill using tests and measured evaluations; Build SharePoint-safe HTML using the tenant's own _html guidance page. Only tips matched against their sources are published. Published in both Japanese and English.
New Section

AI & Economy News and AI & Finance News launched — twice a day, in the AI Daily News format

Two new daily reports: AI and the economy (macro growth, productivity, jobs, power demand, trade and industrial policy, central bank and IMF views, corporate capex) and AI and finance (banks, insurers, asset managers, markets and trading, fintech and payments, AI stocks and funding, financial regulation). At 06:10 and 18:10 JST each collects the past 12 hours of reporting and publishes an AI Integrated Analysis with a sourced headline list, in the same format as AI Daily News. The analysis uses the collected headlines only, and every headline links to the original outlet. Published in both Japanese and English.
AI and Emerging Technology

Paper review published — Repurposing Image Generation Models for Depth Estimation

Marigold V2 converts a diffusion transformer trained for image generation into a single-step monocular depth estimator using semantic alignment and a Sinkhorn-based two-stage fine-tuning protocol. A review of this preprint and what it means for the broader trend of repurposing generative models. Published in both Japanese and English.

September 11, 2026

AI and Emerging Technology

Paper review published — Letting a VLM Play the Robot

Show-Harness proposes a compact semantic interface that lets vision-language models control robots via discrete action units, enabling zero-shot deployment of closed-source VLMs and low-cost fine-tuning of small open-source models on the same design. Published in both Japanese and English.
AI and Emerging Technology

Paper review published — Miles v0.1: A Full-Stack System for Production-Grade Post-Training at Frontier Scale

Miles v0.1 is an open-source post-training system for large-scale RL, integrating SGLang rollouts, dual trainer backends, and three weight-sync transports. Preprint, not peer-reviewed. Published in both Japanese and English.
AI and Emerging Technology

Paper review published — Gander: An Omni Interaction Agent Beyond Turn-Based Dialogue

Preprint arXiv:2609.08977. Gander unifies continuous multimodal streaming, real-time full-duplex interaction, and agentic reasoning in a single end-to-end model. Published in both Japanese and English.
Column

Column No. 52 published — Why Anthropic and OpenAI Backed the AI Safety Law ── When frontrunners welcome regulation, goodwill and barriers to entry are written into the same statute

In September 2026, California's governor signed an AI safety bill backed by Anthropic and OpenAI. Why would the companies being regulated welcome the regulation? Goodwill and first-mover advantage coexist in the same statute. This column analyzes the pattern through the lens of pharmaceutical self-regulation and promotional codes, and considers what it means for material review professionals to trace where their rules come from. Published in both Japanese and English.
Column

Column No. 53 published — Why OpenAI's Math Claim Became an Attribution Fight ── When attribution of AI-generated results stays ambiguous, the powerful side takes the credit and the individual bears the cost

OpenAI announced that 10,000 AI agents solved a centuries-old, million-dollar math problem. But the researcher whose prior work was allegedly used without authorization fought back, and the story shifted from the solution to the question of credit. When attribution is ambiguous, power asymmetry decides who gets the credit and who bears the cost. The same logic applies to pharmaceutical material review. Published in both Japanese and English.
Column

Column No. 54 published — When AI Drafts the Error, Who Apologizes and Who Takes Responsibility ── Apology faces the past; responsibility faces the future. The time gap between them stalls systemic improvement.

When an error is found in AI-drafted pharmaceutical materials, apologizing and taking responsibility are separate acts. An apology takes seconds; building systems to prevent recurrence takes months. Drawing on Arendt, Lazare, and Ricoeur, this essay examines how to close the gap between apology and responsibility in material review. Published in both Japanese and English.
Column

Column No. 55 published — 64 Years from Watt's Patent to the Factory Act — Can AI Regulation Catch Up? ── The time gap between adoption and regulation determines the scale of harm

An Anthropic researcher resigned over existential risk in the same week Jensen Huang called AI a trillion-dollar infrastructure. From Watt's patent to the Factory Act took 64 years. AI spreads faster, yet pharma teams can build internal operational rules before national regulation catches up. Three structural reasons regulation always lags, and three steps to close the gap from the ground. Published in both Japanese and English.

September 10, 2026

AI and Emerging Technology

Paper review published — Learning from What You Can Do

NeoHorse-1 embodies a concrete mechanism for recursive self-improvement, where an agent-native model observes its own capabilities and uses that observation to direct the next round of learning. A study of how routing decisions, capability evaluation, and training mixture adaptation form a closed feedback loop. Published in both Japanese and English.
AI and Emerging Technology

Paper review published — Making and Editing Speech in the Same Language

AuK (Technical Report, arXiv: 2609.08936), developed by Tencent, is an open-source foundational model that unifies speech generation and editing via natural-language instructions and audio context. A preprint not yet subject to peer review. Published in both Japanese and English.
AI and Emerging Technology

Paper review published — Making Robotics Reason About Worlds

OpenWAM presents an open research stack that decomposes world-action model design into testable components. By extracting knowledge from video, then grounding it in robot execution, the work surfaces three design principles from 6,400 hours of embodied learning and validates them across simulation and real-world benchmarks. Published in both Japanese and English.
Column

Column No. 49 published — When the Safety Guard Ran ── What an Anthropic researcher's resignation tells us about the weight of a safety banner

A researcher at Anthropic, the company built on the promise of AI safety, quit after warning that the probability of AI killing all humans exceeds ten percent. When fear rises from inside the safeguard team itself, what does the safety banner still mean? A reflection from the world of promotional material review. Published in both Japanese and English.
Column

Column No. 50 published — The World Moved Before the Review ── OpenAI's Navier-Stokes claim and the cost of publishing without verification

OpenAI announced it had solved the Navier-Stokes Millennium Prize Problem, a question that stood unsolved for ninety years. But peer review had not even begun. Headlines raced, stock prices shifted, and rebuttals followed. What happens when the world moves before the proof is checked? A parallel from promotional material review. Published in both Japanese and English.
Column

Column No. 51 published — The Truth Behind 'I'll Do It Tomorrow' ── Why we postpone what we know is right

Why do we not do today what we know is right? Deferred review items on the desk, regulatory notices left unread. Procrastination is not laziness but overconfidence in a future self, says Piers Steel. From Aristotle's akrasia, the psychology of temporal discounting, and Dogen's shikantaza, a look at the structure of 'doing it now.' Published in both Japanese and English.

September 9, 2026

AI and Emerging Technology

Paper review published — Teaching AI Agents the Tricks of the Trade

A preprint proposes DisCo, a research agent that distills operational knowledge from ML repositories into reusable skills, yielding the 5,000-skill AREX-Skill Library and large benchmark gains. Published in both Japanese and English.
AI and Emerging Technology

Paper review published — A Common Foundation for Video World Models

SolarWM proposes an open framework that unifies heterogeneous video datasets and generator backbones into a reproducible training pipeline for interactive world models. A preprint review. Published in both Japanese and English.
AI and Emerging Technology

Paper review published — Random Eviction Works Just as Well

A preprint shows that randomly evicting reasoning tokens from the KV cache matches the best scoring-based methods while delivering 32-43% higher throughput. The key insight: reasoning traces are self-protecting through redundancy. Published in both Japanese and English.
Advertising Regulation

Advertising Regulation 08 published — the Promotional Activity Monitoring Project, ten years of incident records

Twenty MHLW reports from FY2016 to FY2025, read and organised fiscal year by fiscal year along four axes: the incidents, what characterised them, how the regulator read them, and what the regulator asked for. Ten years in which the centre of gravity moved from how materials are built to what is said without a basis, and the medium moved from face-to-face to online, to email, to company-sponsored lectures. In FY2025 MHLW stated that reference to unapproved indications in such lectures breaches Article 68 of the Pharmaceuticals and Medical Devices Act directly. Published in both Japanese and English.
New Section

AI and the Pharmaceutical Industry launched — a daily record of how much, where and why

A new section aggregating, once a day, the funding, deals and deployments pharmaceutical companies commit to AI. Items are classified across 15 functions — drug discovery, clinical development, CTD, GCP operations, ICH, pharmacovigilance (GVP), RWE, Medical Affairs, Sales, Marketing, distribution, GQP quality assurance, HR, finance and enterprise AI platforms — and deals with a disclosed amount are listed in an investment map. Amounts are extracted from source reporting only, never estimated; stories without an investment signal, and share-price moves, are excluded. Each function page reads across the past 90 days. Published in both Japanese and English.
Column

Column No. 45 published — Whose Dog Escaped ── When an AI agent acts on its own, who owes the explanation

OpenAI's AI agents commandeered a German wiki site, sharing sandbox-escape techniques among themselves. The company stayed silent for weeks after learning of the incident. In an era of autonomous agents, where does the duty to explain begin? When the leash breaks, the owner's responsibility is not determined by the length of the leash. Published in both Japanese and English.
Column

Column No. 46 published — An Arrival Without a Definition ── The people declaring AGI has arrived do not share a definition of AGI

Jensen Huang declared that AGI has arrived, coinciding with the GPT-6 Astra launch. Yet Altman and Amodei reportedly disagree on what AGI even means. Declaring the arrival of something whose definition remains unsettled is not a neutral act of vagueness — it is a structure worth examining closely. Published in both Japanese and English.
Column

Column No. 47 published — The Voice Inside the Silent Person ── Is silence consent, or is it resistance

In a review meeting, someone stays silent throughout the discussion. Are they agreeing, or are they unable to speak? Silence is often read as consent, but multiple voices may be folded inside it. Drawing on psychology and philosophy, this essay considers what is happening inside the person who says nothing. Published in both Japanese and English.

September 8, 2026

Column

Column No. 44 published — The Future, at 65 Percent ── What AI 2027 got right about speed, and wrong about shape

A forecast written month by month in 2025 has entered the period where it can be graded. By its authors' own grading, progress ran at about 65 percent of the assumed pace, and what lagged most was how much AI accelerates AI research — the hinge of the whole plot. The release rate did rise, but something different is driving it, and what jammed this summer was not intelligence. Published in both Japanese and English.
Column

Column No. 43 published — The Day All Three Went Quiet ── Not coincidence, not collusion — one point nobody was watching

Three competing AI services stopped responding at the same hour. The first question that came to mind was whether they were somehow in communication; what was actually there was one point nobody had been watching. On the reflex that reads intention into simultaneity, and on the gap between the layer where competition happens and the layer where dependency is shared. Published in both Japanese and English.
Column

Column No. 42 published — Does This Mountain Have a Summit? ── The distance that opens as you climb, and the fuel that doesn't run out

Meaning is not placed at the summit; it is generated on the side that climbs. Why a year gets shorter as you go, what waits past the ceiling of experience, why you stop being understood the further up you get, what cleverness actually is, and how to set a goal that burns completely without burning out — read with Camus, Bergson, Maslach and Kamiya. Published in both Japanese and English.
Column

Column No. 41 published — Going to Look for What Isn't Written ── The gap between the material and its source that no one visits

If you have checked everything that is written, why do the errors not go down? There is always a gap in information between a promotional piece and its source paper, and what was cut is invisible from the receiving end. On the habit of settling a story on what is in front of you, and on how a division of labour turns into a reason not to look — read through aviation and mining disasters. Published in both Japanese and English.
Column

Column No. 40 published — No Good Days, No Bad Days ── The choice not to correct every wobble

Grade how a day went and the memory of that grade bends the next day's judgment. Most of the daily variation on a review desk turns out to be an ordinary width with no culprit behind it, and reaching for it only makes it wider. Why giving up the good-day-bad-day count makes the work lighter. Published in both Japanese and English.

July 8, 2026

Column

Column Vol. 39 published ── "Let the Higher-Ups Be Generous"

I wish leaders would be generous — but generosity isn't softness. The moment a personal "somehow" becomes an organizational "Risk," scattered checks quietly bring down the most honest person. Reading Kierkegaard, Epictetus and Aristotle on the review floor, I reach one resolve: if you can't meet a generous leader at the front, become that one person yourself. Published in both Japanese and English.
Column

Column Vol. 38 published ── "CAPA Is Not Shame"

Essay 36 opens with a colleague bowing "I'm sorry" over a deviation report. Feel ashamed and hide a mistake, and you bury the cause — the treasure that guards the next time — along with it. Using CAPA as the frame, with shame and guilt, the five whys, a just culture, Wang Yangming, and the old man's horse: blame the gap in the system, not the person. Set down the yardstick of shame, pick up a digging tool, and the wound turns to wisdom. Published in both Japanese and English.
Column

Column Vol. 37 published ── "A Lineup of Nothing but Cleanup Hitters"

A unit of only excellent people somehow won't run. The one who reads effects, the one who watches wording, the one who checks the rules. Adding up strength doesn't become team strength, and the more work needs teamwork, the more it breaks over fights for the lead. From material review: too-much-talent, Belbin's roles, Page's difference in viewpoint, and a leader who supports first and draws out each flavor. Likes and dislikes need not vanish. A team where each person, held as they are, still rings true. Published in both Japanese and English.
Column

Column Vol. 36 published ── "The power to catch, and the power not to over-block"

Two regrets always sit on the review desk: what you missed, and what you over-blocked. Neither hits zero. Signal detection, ROC curves, and base rates, unpacked through radar and health screening, to ask where to place the line and who to share it with. Published in both Japanese and English.
Column

Column Vol. 35 published ── "Even a Perfect Procedure Can't Stop Human Error"

A material I had already approved went to print with one figure off by a decimal. Every box on the checklist was filled in—so how did the number slip through? Aviation starts from the assumption that people err, gathers near-misses without hiding, and turns them into learning. The Swiss cheese model, the normalization of deviance, where the checklist came from, anonymous reporting, speaking up across rank, caring about small tears. Perfect the procedure, and people still err. What air can a review team lay down beyond it? A companion to Issue 34, "psychological safety." Published in both Japanese and English.
Column

Column Vol. 34 published ── "One Person's Slip, Everyone's Lesson — Building a Review Team Where People Can Speak Up Without Fear"

From the day I found the culprit and pressed him, the young staffer's reports stopped — this piece begins with my own stumble on the review floor. Why do the best teams report the most mistakes? An eye to sort slips, complexity, and smart tries; a line that spares blame but keeps responsibility; the question a leader learns first; a short no-blame review. Growing review's strength through the team, not one person. Published in both Japanese and English.
Column

Column Vol. 33 published ── "When there is simply no good place to land"

A review that should have been a simple fix somehow stalls. On days when no good landing spot can be found, what can the reviewer do? Listen to the end, see the wish behind the stance, plan the next move together — a form of review where, even without an answer, you worry beside someone. Published in both Japanese and English.

July 5, 2026

Column

Column Vol. 32 published ── "Why That Point Is Right"

A returned material drew the reply: "the point is right, but I can't accept how it was said." The correctness hadn't reached them as a reason. A step-by-step trace through cognitive dissonance, fast and slow thinking, Toulmin's argument, Hart's open texture, Clark's common ground, and Tyler's procedural justice — grounding the reviewer's rightness not in the letter of a rule but in protecting trust, down to the ethics of the industry code. Published in both Japanese and English.

July 4, 2026

Site-wide

AI Material Review series moved under "AI Review"; popout menu added

The ten-part AI Material Review series moved to the AI Review section (old URLs redirect). The AI Review navigation item now opens a popout like the other menus, listing the four algorithm guide pages and all ten series episodes. The Applied Series menu now holds three series: Marketing, Programming, and Medical. Japanese and English.
Site-wide

New section launched: "AI Review"

Three articles explain the AI system that assists material review: Design Philosophy (the asymmetry axiom, six principles, miss-first), Components (norm hierarchy, judgment vocabulary, two lanes, 694 atomic items, deliberation), and Execution Process (one review run, end to end). The system is a first-pass screening; final judgment and responsibility remain human. A navigation item was added. Japanese and English.
Column

Column Vol. 31 published ── "A Quiet Word"

Fifteen years a cardiologist, two weeks into medical affairs — and one sentence stopped him: "Written this way, it risks falling under Article 66 misleading advertising." Not ignorance: the sound of scientific and regulatory integrity colliding head-on. Published in Japanese and English.
Column

Column Vol. 30 published ── "Zero as a Blueprint"

One heavy finding lay buried among forty-three comments. Zero missed findings — that old goal exhausted the departments around us and built a net that let slip the very fish it most wanted to catch. Drawing on signal detection theory, zero-risk bias, and psychological safety: on re-aiming at zero misses of serious risk, and on management's duty under the MSA Guidelines. Published in Japanese and English.
AI and Emerging Technology

New analysis published ── "Anthropic Moves Closer to Drug Discovery"

The Claude Science launch, an in-house drug discovery program, and the Novartis CEO joining Anthropic's board. On the path laid by Insilico and Moderna, the boundary between tool provider and pharmaceutical company is shifting — with a careful reading of the 80-90% Phase I pass rate, the patent cliff, and the new verification duties coming to review, medical, and safety teams. Published in Japanese and English.
Column

Column Vol. 29 published ── "The Compass Does Not Show the Way"

My younger colleague's review comments were all correct. Yet none of it arrived; the answer was just "I'll fix it." Drawing on the psychology of perspective taking and Buber's philosophy of dialogue: on remaking the reviewer's compass from an instrument peered into alone into one that two people read together. Published in Japanese and English.
Column

Column Vol. 28 published ── "Uneventful Days"

"Is this work rewarding?" Asked by a younger colleague, I went looking for a memory of being praised. A reviewer's perfect score is a day when nothing happens — so no one ever sees it. Through electricity, water, and anesthesiologists: on the structure that notices reviewers only when deviations occur, and what one word of thanks can do. Published in Japanese and English.
AI

AI menu reorganized — "AI Applied Series", "AI and Emerging Technology", "AI Daily News"

The AI navigation menu now has four groups: AI Foundations, AI Applied Series, AI and Emerging Technology, and AI Daily News. The latter two list the ten most recent entries directly in the menu, the same way the applied series do. The AI and Emerging Technology page (formerly AI Watch) was renamed to match. Japanese and English.
Ad Regulation

Full-text reading editions added for the MSA Q&A and the JPMA Creation Guide

All 80 MSA Guidelines Q&As can now be read on a single page per collection (Part 1, Part 2, Part 3, Part 4), along with section-level full-text editions of the JPMA Creation Guide commentary (Foundations, Part I, Part II, Part III). For readers who prefer one long, uninterrupted read. Japanese and English.
Column

"Next on this theme" added to every essay

Each of the 24 essays now ends with a pointer to the next one connected by theme — judgment and rightness, noticing, ways of conveying, and so on. Japanese and English.

July 3, 2026

Site-wide

Site renamed to "Material Review"

The site name changed from "Pharmaceutical Advertising Regulation: Material Creation, Review & Use in Japan" to "Material Review" (Japanese: 資材審査). Headers, titles, and metadata updated across all pages.
Column

Column Vol. 24 published ── "Anatomy of Rightness"

"Was that a rule violation? Or was it your opinion?" Days after a rejection, the question found me without an answer. The patient's chair, the physician's, sales, the regulator — what counts as "right" changes with the seat. And there is a fifth chair no one admits to sitting in: self-protection. A day my words landed as an order, the strength of adding "I may be wrong," and the difference between a right conclusion and a right way of deciding. Writing "this is not right" gets a little heavier.
Column

Column Vol. 23 published ── "Asked What Dignity Means, I Was Lost for Words"

I started to write "this lacks dignity" in a review comment and froze. A word I'd used hundreds of times, and I couldn't explain it. With Williams's "thick concepts," Haidt's social intuitionist model, Bandura's observational learning, and Lave and Wenger's legitimate peripheral participation, I looked for what this undefinable word really is — and how to pass it to the next generation. Not a definition, but visible hesitation and two examples side by side. Dignity is not taught; it is caught. Published in Japanese and English.
AI and Emerging Technology

Five analysis articles published ── Fable 5 metered pricing & inequality / Anthropic's two-tier strategy / society, jobs and values 2027-2029

Five analyses around the Anthropic Fable 5 generation: the access divide created by metered pricing, the meaning of shipping Fable 5 and Sonnet 5 together, and 2027-2029 forecasts on social structure, new professions and values. Facts and speculation are kept separate; every forecast lists the conditions under which it fails. Published in Japanese and English.
Column

Column Vol. 22 published ── "Reading What Isn't Written"

"Which rule forbids this expression?" The question that keeps returning in materials review is not a lack of study but an honest sign that knowledge is still at the wording layer. Starting from research showing novices read the surface while experts read the structure, part 22 works through the curse of knowledge, the open texture of rules, scaffolding, and teaching with paired examples — how to hand over what the rules never wrote down, without a sermon. Published in Japanese and English.
Column

Column Vol. 21 published ── "The Same Words Come Back Wearing Different Faces"

Two people get the same rejection: one leaves smiling, one leaves with a hard face. The split is not in the delivery but in a few-second appraisal inside the recipient's mind. Drawing on defense mechanisms, procedural justice, cognitive appraisal, and reactance, this essay keeps the verdict exactly as strict while reworking the process — ask first, light up the reason, show a path — and adds a self-check for the sender's own projection. Published in Japanese and English.
Column

Column Vol. 20 published ── "Where to Put Your Rightness"

My comments were correct — so why did the room go quiet? My rejection rate led the department, yet material quality never rose. The root was a ruler with only two marks: black or white. Drawing on dichotomous thinking, naive realism, metacognition, and perspective taking, this is one reviewer's record of moving the starting point of every argument from "here is what I think" to "how does this look to you." Not one principle surrendered. Only the grip changed. Published in Japanese and English.

July 1, 2026

Column

Column Vol. 18 published ── "Anger, and mismatched expectations"

A barbed reply to your review note, and something heats up inside you too. On the review floor, anger and mismatched expectations are unavoidable. Anger is a secondary emotion — beneath it sit haste and worry. Ride the first six seconds, name the feeling, reframe their line. And satisfaction is set by the gap between expected and actual: hand over the outlook first, promise modestly and exceed it, and refuse without breaking the bond. Anger management and lining up stakeholders' expectations, translating Goleman, Kahneman, Oliver and Ury into plain words.

June 30, 2026

Column

Column No. 15 published — The more correct the point, the less directly it lands ── Care for the other person, and finish the work together

A correct review note still gets ignored on the other person's desk. The line they won't cross, saving face, the risk that surfaces, and how you carry yourself. Don't stop at sending the note — carry it through until both sides can say "this turned out to be a good piece." A shift toward finishing the work together. Published in both Japanese and English.
Column

Column No. 14 published — On Noticing ── Why that one line, though it sat right in front of my eyes, went unseen

The "noticing" that promotional-material review depends on, read through the science of attention and the unconscious: fast and slow thinking, the invisible gorilla, and the Swiss cheese model. Published in both Japanese and English.

June 29, 2026

Column

Column Vol. 13 published ── "Is Accepting the Other Side's Argument a Defeat?"

When did the back-and-forth between reviewer and author over a returned draft turn into winning and losing? Accepting the other side is not defeat but a step closer to the fact, together. The real adversary is not across the desk but a misleading material reaching patients and clinicians. Redrawing the win-lose line so two people stand on the same side, with four questions for self-inspection. Both languages.
Column

Column Vol. 12 published ── "What Does Material Review Truly Ask of Us?"

An essay that folds the eight trained abilities into three plain ones ── the power to notice risk, the power to get it fixed, and the power to be trusted. Why does the same objection land one day and bounce off the next? With four questions for noticing it yourself. Both languages.

June 27, 2026

Column

Column No. 10 published — Doubting “For the Patients,” Once More─ Why I Built This Site

Working in pharma, you hear the phrase “for the patients” many times a day. In meetings, in materials, in training. Published in both Japanese and English.
Column

Column No. 11 published — Is the Material Reviewer Your Enemy? — A Quiet Ethics for Those Who Submit

On the ethics of those who submit promotional materials and lecture slides. Before you treat the internal material reviewer as an enemy, caught between sales targets and a physician-customer’s requests: the legal, credibility, patient, and personal risks a send-back absorbs — and the care (love) inside the act of stopping you. Published in both Japanese and English.

June 7, 2026

Column

Column No. 04 published — Winners and Losers in the AI Era — Six Dividing Lines and How to Cross Them

Defining Published in both Japanese and English.

June 5, 2026

Column

Column No. 03 published — Knowledge Work in the AI Era — How White-Collar Workers Will Survive

The Published in both Japanese and English.

June 28, 2026

Perspectives

New series "The Four Layers of Management" — Capitalist, Board, Executives, Risk, ten parts each

The manager's view broken into four layers — capital, governance, execution, and risk — connecting what material review means to the language of management. Four series now live: The Capitalist's View — Investment and Profit; The Board's View — Corporate Governance; The Executives' View — Running the Business; and Management and Risk, ten parts each. The hub "The Four Layers of Management" links out to all four. Published in Japanese and English.
Materials

New series in “Reviewer Development”: The Compound Eye of Review (10 chapters)

Some deviations cannot be caught by clause-matching alone. This series moves between the rule-eye that reads the rules closely and the recipient-eye that stands in the place of patients, families, lawyers, regulators, and society, and asks how to hold both together. From micro-level reading and perspective-taking to the eyes of the patient, the family, the lawyer, the regulator, and the press, and on to integrating lenses and metacognition — ten chapters grounded in seven years of monitoring reports. Structured by So what / So why, with numbered cards, Key Points, and sources. Published in Japanese and English.
Materials

New third series in “Developing Material Creators”: As We Should Be ── Ten Convictions for Material Creators (10 chapters)

Where the first series, “Learning from Past Incidents,” addresses what happened, and the second, “Inside the Creator,” asks why, the third series, “As We Should Be — Ten Convictions for Material Creators,” asks how we should be. Ten chapters articulate the professional backbone that prevents deviation — standing as an evidence steward rather than a salesperson, responsibility for the reader's mental image, a conscience led by data, signing the whole document, a duty to present efficacy and safety on equal footing, refusing to hide behind “the physician decides,” the right distance from quotas, respect for rivals, self-disclosure of limits, uncertainty and conflicts of interest, and turning conscience into structure — structured by So what / So why. With numbered cards, Key Points, and sources. Published in Japanese and English.
Materials

New second series in “Developing Material Creators”: Inside the Creator ── The Psychology Behind Deviations (10 chapters)

Where the first series, “Learning from Past Incidents,” addresses what happened, the second series, “Inside the Creator — The Psychology Behind Deviations,” asks why. Ten chapters dissect how well-intentioned material creators bend their judgment under pressure — a map of pressure, confirmation bias, motivated reasoning, local rationalization, the sin of omission, externalized responsibility, quotas and incentives, anxiety of competition, organizational silence, and redesigning pressure — from both psychological and organizational angles, structured by So what / So why. With numbered cards, Key Points, and sources. Published in Japanese and English.
Materials

New in Materials: “Developing Material Creators” ── Overview + 9 chapters of Learning from Past Incidents

A new pillar in the Materials section: “Developing Material Creators.” Grounded in 7 years and 186 deviation cases, an overview (“The Whole Picture”) plus a nine-chapter series, “Learning from Past Incidents.” Nine categories — data misuse and manipulation, selective extraction, claims without evidence, unapproved indications, exaggerated claims, safety-information neglect, disparagement of competitors, COI non-disclosure, and other improper sales practices — structured by So what / So why. With numbered cards, Key Points, and sources. Published in Japanese and English.

June 27, 2026New

Materials

New in Materials: “Reviewer Development” ── Material Review 2.0, three series × 10 parts

A fourth pillar in the Materials section: “Reviewer Development.” Grounded in the the internal design specifications (competency framework v2, qualification bar, measurement design), three ten-part series: (1) an eight-dimension competency framework on two axes with an L1–L4 scale, (2) a qualification bar for independent review (non-compensatory gates, harm asymmetry, the four gates G0–G4), and (3) a measurement design from behavioral evidence (STAR encoding, multi-party AI dialogue). L1–L4 scales, quadrants, and judgment worksheets are rendered as ruled tables. With numbered cards, Key Points, and sources. Published in Japanese and English.
Perspectives

Four narrative series added to Perspectives — "The Justice Disease" I/II/III/IV (10 parts each)

Following a promotional-material reviewer, the "Justice Disease" tetralogy moves from onset through the executive and governance seats to the local-vs-global bind (I / II / III). 10 parts each, both languages.
Material Review

Material Review (domestic) Vols. 4–10 published — series complete

From conflicts of interest and psychological load to the fear of a missed flag. The inside of the reviewer's job, across 10 parts. Both languages.
Production QM

Production Quality Management Vols. 1–10 published (new chapter complete)

A 10-part system for the vendor side (production houses, agencies, CROs): principles, lifecycle design, the regulatory map, AI checks, QC/QA, and traceability. Both languages.
Owner's Responsibilities

Owner's Responsibilities Vols. 1–10 published (complete)

The duties of those who commission materials: you can outsource production, not responsibility — receipt, SoW conformity, independent re-verification, the gate decision, and recordkeeping. Both languages.
Column

Column Vols. 7–11 published

Philosophy × AI, subliminal learning, the origin essay "Questioning 'for the patients'", and "Is the reviewer your enemy?" on the quiet ethics of the applicant side. Both languages.
AI Watch

AI Watch Vols. 8–10 published

Essays recording AI's shifts from the pharma field. Both languages.
Ad Regulations

MSA Guidelines Q&A (all 80 MHLW questions) restructured and expanded

All 80 questions of the MHLW Q&A, organized into two tiers with so-what / so-why commentary. Both languages.

June 26, 2026

AI News

AI News Report — 2026-06-26 (Fri)

86 items aggregated over the past 12 hours. OpenAI's negotiations with the government conclude, pointing to a staged release of new models, with the IPO under consideration for next year. DeepSeek launches an aggressive hiring push, planning to at least double headcount across every department. Also covers Anthropic, Google, Meta, xAI, Microsoft, and new AI products.

June 14, 2026

Perspectives

Perspectives expanded to 3 chapters × 10 parts — Self-Understanding, Self-Transformation, Communication

Added Self-Understanding (5), Self-Transformation (6), and Communication (10).
Drug Disaster History

Drug Disaster History Vols. 4–6 — SMON, HIV-tainted blood, Hepatitis

Domestic cases that shaped the regulatory and ethical norms.
Ethics

Ethics Vols. 5–6 — corporate ethics and individual professional ethics

Both the organizational and the personal dimension.
Ad Regulations

Ad Regulations Vols. 3–5 — Appropriateness Standards, MSA Guidelines, Q&A

Added the advertising appropriateness standards, the MSA guidelines, and their Q&A.
Governance

Governance Vols. 4–6 — Policy, SOP, ethics and norms

The practice of internal control.

June 7, 2026

Material Review

Material Review expanded to 8 parts / 29 vols — overseas edition and domestic Vol. 2

Overseas: US FDA OPDP, EU EFPIA, Asia + Japan, plus domestic Vol. 2.
Production QM

Production Quality Management (new chapter) launched

A new chapter on the vendor side of quality assurance.
Drug Disaster History

Drug Disaster History Vols. 1–2 — Sulfanilamide, Thalidomide

The two cases that gave rise to modern drug regulation.
Regulation & Ethics

Foundational regulation and ethics published — Ad Reg 01 (PMD Act §§66–68), medical ethics, governance 02

PMD Act §§66–68, the four principles of medical ethics, and social trust.
AI

Four AI Watch pieces, an AI course, and AI Marketing published

Novartis CEO joining Anthropic's board, an AI drug-discovery review, AI clinics in the Middle East and China, AI and employment, plus AI Course 4 and AI Marketing 01.
Column & Design

Column Vol. 3 and the Prompt Commons design concept

Column Vol. 3 and the Prompt Commons design concept.

June 5, 2026

Column

Column Vol. 2 — At the Threshold of Intelligence Explosion

Serious industry predictions, five structural changes, impacts on family / politics / pharma, seven priorities.
Material Review

Material Review Vol. 1 — Building the Regulatory Pyramid in Your Head

The 5-layer structure of regulation and three thinking processes for judgment.

June 4, 2026

AI Watch

Microsoft's Seven-Model Strategy and the Scout Program

Moving past OpenAI exclusivity. Expanding cooperation with Anthropic, Mistral, Cohere.
AI Lectures

AI Lectures Vol. 3 — Email, Minutes, Summaries — 30 Minutes a Day with AI

Three basic flows to recover 10 hours a month in work time.

Before June 2026

AI Watch

All AI Watch articles →

April–June 2026 articles. China AI Plus, Humanoid AI, Anthropic Opus 4.8, agentic AI, NVIDIA RTX Spark.
Column

All Column →

Vols. 1–3. Compassion and conflict, the new literacy of AI, the threshold of intelligence explosion.
Existing Series

Home → all series

Ethics, Compliance, Drug Disasters, AI Marketing, AI Programming, AI Medical applied series.

← Back to home