Regulator probe and voluntary pledge on one day── FTC probe, leaders' voluntary standards, state-aware defense
Download the video (MP4, 12 MB)
On the same day, two moves pointed in opposite directions. A US agency began examining the people who build AI. At the same time, the government and those same builders agreed to police themselves. Will rules bind from outside, or will firms be trusted to guard from within?
Nobody knows yet which path will win. Meanwhile, AI work in pharma grows every day. Users cannot wait for the rules to settle. On what basis should they choose tools, and how much should they hand over?
I look for answers in shifting rules and money flows. I also look at the code under the tools, and at how records are kept.
01FTC investigates leading AI firms
Source New York Post ／ cnbc.com
This morning's top story: AI builders are now the ones under scrutiny. A government body has started asking whether their tools hurt the people who use them. That matters for users like us, too.
| Item | As reported |
|---|---|
| Agency | FTC (US Federal Trade Commission) |
| Targets | Developers of 'super intelligence' models incl. Anthropic, OpenAI |
| Focus | Potential consumer harm and product risks |
| Reported by | Reuters, CNBC, WSJ, NYT, CBS and others |
| Scale | New York Post exclusive: a 'sweeping probe' |
Reported 09/30. Our site's analysis is based on headlines only
This changes who judges the danger. Until now, builders described the dangers in their own words. From here, an outside body gathers evidence and decides. And the worry is not a distant future. It is harm to people using these tools today. Tools called clever must now pass the safety test any product faces. Many outlets ran the story at once, so I expect it to last.
This body's original job is to protect buyers. With it stepping in, AI danger becomes a question about the deal between maker and user. I think the era of simply trusting the builders' goodwill is nearing its end.
For pharma, this is not a faraway story. If an in-house tool's maker comes under review, your terms could change midstream. Which tool, for which task, checked in which way? Companies that keep that record stay calm when terms shift.
Yet on the very day the review began, the builders were elsewhere, promising to keep themselves in check.
02A voluntary pact on the same day
Source TechCrunch ／ The New York Times ／ Futura, le média qui explore le monde ／ Nautilus | Science Connected
So on one day, the government set out two opposite approaches. One restrains from outside. The other asks firms to hold back from within. Which one leads will shape how users prepare.
| Axis | Voluntary AI standards | FTC probe |
|---|---|---|
| Actors | President Trump and top AI executives | FTC |
| Form | Agreed at a summit; pledge document signed | Sweeping probe of product risks |
| Binding force | Doubts remain over binding force | Proceeds in parallel as law enforcement |
| Around it | NYT opinion questioned letting AI giants self-regulate | |
The pledge reportedly misspelled 'United States' (TechCrunch)
Side by side, the gap is about penalties. The signed promise does not say what happens if broken, so people question its force. The review, by contrast, can end in enforcement. Running a weak promise and a real review together suggests the government has not chosen a course. The signed paper even carried a basic error, which undercut it on day one.
- Role
- Nobel Laureate
- Remark
- Warned extinction risk of '10% is not unreasonable'
- Role
- NVIDIA CEO
- Remark
- Told AI labs they could stop if concerned
The voluntary safety pledge, called 'binding', was reportedly mocked
Those who built AI also disagree. A veteran researcher warned against taking the danger of our species dying out lightly. The head of a chipmaker told worried labs they could simply quit. If builders cannot agree on the danger, each signer's promise will mean something different.
Pharma has long relied on both law and industry codes to keep its conduct in line. People in the field know neither is enough alone. So companies set the limits for in-house AI use now, without waiting for outside rules.
While rules wavered between promise and review, nations were also drawing lines in the code that powers AI.
03Partnering on software that runs chips
Source The New York Times ／ Reuters ／ Seeking Alpha
Now we move from rules to what is inside the tools. The rivalry between nations now reaches past visible hardware. It extends to the invisible code that makes that hardware run.
The target is not the hardware itself. A processor only performs well when paired with the code that runs on it. Reports say one firm's lead lies partly in that code. I think that is because so many developers are used to it. Rebuild the same code in a form anyone can use, and moving to other hardware gets easier. The partners want to cut that effort and stop depending on a single supplier.
That strength did not come from speed alone. Over many years, the company drew developers' habits and their stored programs to its side. If that bond weakens, the reasons for picking one processor over another change too.
Pharma research and analysis run on the same code and hardware. If these split by nation, trade rules may choose your tools instead of performance. Firms running the same analyses at overseas sites are the most exposed.
As nations draw lines in code, large sums aimed at health care also moved, outside the drug companies.
04$500 million for a healthcare AI lab
Source The Pharma Letter
Where money goes shows where people expect growth next. Today's largest sum went not into a drug company, but into a new group formed outside one.
New healthcare AI lab
Executives from biopharma, Meta and AWS launched Ortet, a lab for health and medical AI.
An amount as a promise
Headlines say "backing" and "commitment". Not readable as money already spent or a fixed investment.
What is unknown
Backers and planned uses are unclear from headlines. No pharma partnership is indicated either.
Its founders worked on AI at drug makers and at big technology firms. They set up a health-focused research group outside any one company. So drug makers are no longer the only drivers of medical AI. But who is paying, and for what, is not clear from this report. Nor do we know whether it will team up with drug makers. The size arrived before the substance.
Also, the sum is a pledge to pay, not money already paid. Pledged amounts can shrink as conditions change. I think it is too early to say money is pouring into this field based on a headline figure. Whether the pledge is real will show only when payment records appear.
Pharma professionals see daily the gap between announcing a development plan and winning approval. Funding promises need the same caution. Check who is paying, and when it was paid, before treating them as a partner.
From money, we turn to AI that carries out actions itself. There, a danger that is easy to miss was waiting.
05Stopping steps that each look harmless
Source SEAD: A State-Based Perspective on Attack and Defense in Tool-Using Agents ／ Daily Papers: SEAD
AI that acts on its own is entering the workplace. The question is how to tell which action is dangerous, when, and by looking at what.
The path the study describes is simple. The first request widens who can view a document. The next makes an ordinary duplicate. Each request alone is harmless. Together, though, an internal draft becomes readable from outside. The danger was never in the words. It sat in the condition left behind by the earlier action.
| Axis | Judge by wording | Judge by checking state |
|---|---|---|
| What it looks at | Wording of instructions or proposed actions | Current state of files and permissions touched |
| How it checks | Looks for danger signs in the wording | Confirms via read-only queries before execution |
| When split into harmless steps | Finds no reason to stop at any point | Judges by overlap with state left by earlier steps |
| Publication | arXiv preprint; not peer reviewed | |
A defense that reads only each request's words cannot stop this. Taken one by one, no request gives a reason to halt. Look at the current settings just before acting, and the overlap shows. The defender's focus shifts from words to conditions. But experts have not yet reviewed this work. Others still need to test it.
Pharma holds many documents that must never leave the company. Think of pre-approval files and patient records. If AI acts for you, something must check the viewing settings before each action. Reading logs afterward only reveals a leak once it has happened.
The idea of checking conditions also links to efforts to record where AI output comes from. Those efforts are happening right beside our own work.
06Two moves close to pharma practice
Source Google DeepMind ／ Nature ／ Morningstar
Things AI produces often cannot be traced later. Who made them, and how? Where and how to keep that origin is the next question.
| Axis | Protein watermarking | Data foundation |
|---|---|---|
| Who | Google DeepMind | A pharma company |
| Target | AI-generated proteins | Master data |
| What | Watermarks that preserve function | Global rollout of Veeva OpenData |
| Positioning | Released as SynthID Bio; paper in Nature | Data first, as groundwork for scaling AI |
One effort places a mark inside a substance that AI designed. If the mark goes in without harming how the substance works, its origin can be traced. The other is a drug maker tidying its core records before spreading AI across the company. It chose to fix the data underneath before the models on top. Both aim to keep outputs and inputs explainable later.
A mark helps only if it leaves the contents working as before. If it weakened the result, nobody would use it. Showing that it need not do so makes origin tracking a more practical choice for users.
Pharma has always earned trust through records. If you cannot show who checked what, and when, a submission will not be accepted. That principle holds with AI too. How complete your origin records and core data are will set how much work you can hand to AI.
Apart from tidying origins and records, though, some questions still have no answer from anyone.
07What is being overlooked
Source NBC News ／ dailycampus.com ／ Tom's Hardware
Outside the big headlines, several things remain unsettled. Listing what is unknown also shows what today's big stories are missing.
Skipping a hearing
Sam Altman is expected to skip a congressional hearing on rogue AI agents, reports said (NBC).
Law in force from Oct. 1
New AI regulation law in force from Oct. 1 was reported. Which law, and its details, are unclear from headlines.
Chinese model's capabilities
Anthropic claimed that a popular Chinese AI model has "Mythos-class" hacking abilities.
One major company head is expected to stay away when lawmakers question risky AI. Avoiding that while under review could shake trust in the company. A company under scrutiny needs to explain itself in public. Another statute, said to apply from today, was also reported. Which one, and what it says, is still unclear. One developer claimed a rival nation's well-known system can mount strong attacks. Yet here, no outside party has confirmed that claim. Claims about a competitor are safer left out of decisions until evidence appears.
Recording the unknown as unknown is also a core skill in practice. Keep verified and unverified reports apart, and later judgments go wrong less often. Companies that note open questions can update internal rules fast once a rule's content is clear.
So, with the certain and the uncertain kept apart, what can users actually rely on?
The scope of the FTC probe, and how each targeted firm responds.
Open the full transcript
Intro
On the same day, two moves pointed in opposite directions. A US agency began examining the people who build AI. At the same time, the government and those same builders agreed to police themselves. Will rules bind from outside, or will firms be trusted to guard from within?
Nobody knows yet which path will win. Meanwhile, AI work in pharma grows every day. Users cannot wait for the rules to settle. On what basis should they choose tools, and how much should they hand over?
I look for answers in shifting rules and money flows. I also look at the code under the tools, and at how records are kept.
CH 01 FTC investigates leading AI firms
This morning's top story: AI builders are now the ones under scrutiny. A government body has started asking whether their tools hurt the people who use them. That matters for users like us, too.This changes who judges the danger. Until now, builders described the dangers in their own words. From here, an outside body gathers evidence and decides. And the worry is not a distant future. It is harm to people using these tools today. Tools called clever must now pass the safety test any product faces. Many outlets ran the story at once, so I expect it to last.This body's original job is to protect buyers. With it stepping in, AI danger becomes a question about the deal between maker and user. I think the era of simply trusting the builders' goodwill is nearing its end.For pharma, this is not a faraway story. If an in-house tool's maker comes under review, your terms could change midstream. Which tool, for which task, checked in which way?
Companies that keep that record stay calm when terms shift. Yet on the very day the review began, the builders were elsewhere, promising to keep themselves in check.
CH 02 A voluntary pact on the same day
So on one day, the government set out two opposite approaches. One restrains from outside. The other asks firms to hold back from within. Which one leads will shape how users prepare.Side by side, the gap is about penalties. The signed promise does not say what happens if broken, so people question its force. The review, by contrast, can end in enforcement. Running a weak promise and a real review together suggests the government has not chosen a course. The signed paper even carried a basic error, which undercut it on day one.Those who built AI also disagree. A veteran researcher warned against taking the danger of our species dying out lightly. The head of a chipmaker told worried labs they could simply quit. If builders cannot agree on the danger, each signer's promise will mean something different.Pharma has long relied on both law and industry codes to keep its conduct in line. People in the field know neither is enough alone. So companies set the limits for in-house AI use now, without waiting for outside rules. While rules wavered between promise and review, nations were also drawing lines in the code that powers AI.
CH 03 Partnering on software that runs chips
Now we move from rules to what is inside the tools. The rivalry between nations now reaches past visible hardware. It extends to the invisible code that makes that hardware run.The target is not the hardware itself. A processor only performs well when paired with the code that runs on it. Reports say one firm's lead lies partly in that code. I think that is because so many developers are used to it. Rebuild the same code in a form anyone can use, and moving to other hardware gets easier. The partners want to cut that effort and stop depending on a single supplier.That strength did not come from speed alone. Over many years, the company drew developers' habits and their stored programs to its side. If that bond weakens, the reasons for picking one processor over another change too.Pharma research and analysis run on the same code and hardware. If these split by nation, trade rules may choose your tools instead of performance. Firms running the same analyses at overseas sites are the most exposed. As nations draw lines in code, large sums aimed at health care also moved, outside the drug companies.
CH 04 $500 million for a healthcare AI lab
Where money goes shows where people expect growth next. Today's largest sum went not into a drug company, but into a new group formed outside one.Its founders worked on AI at drug makers and at big technology firms. They set up a health-focused research group outside any one company. So drug makers are no longer the only drivers of medical AI. But who is paying, and for what, is not clear from this report. Nor do we know whether it will team up with drug makers. The size arrived before the substance.Also, the sum is a pledge to pay, not money already paid. Pledged amounts can shrink as conditions change. I think it is too early to say money is pouring into this field based on a headline figure. Whether the pledge is real will show only when payment records appear.Pharma professionals see daily the gap between announcing a development plan and winning approval. Funding promises need the same caution. Check who is paying, and when it was paid, before treating them as a partner. From money, we turn to AI that carries out actions itself. There, a danger that is easy to miss was waiting.
CH 05 Stopping steps that each look harmless
AI that acts on its own is entering the workplace. The question is how to tell which action is dangerous, when, and by looking at what.The path the study describes is simple. The first request widens who can view a document. The next makes an ordinary duplicate. Each request alone is harmless. Together, though, an internal draft becomes readable from outside. The danger was never in the words. It sat in the condition left behind by the earlier action.A defense that reads only each request's words cannot stop this. Taken one by one, no request gives a reason to halt. Look at the current settings just before acting, and the overlap shows. The defender's focus shifts from words to conditions. But experts have not yet reviewed this work. Others still need to test it.Pharma holds many documents that must never leave the company. Think of pre-approval files and patient records. If AI acts for you, something must check the viewing settings before each action. Reading logs afterward only reveals a leak once it has happened. The idea of checking conditions also links to efforts to record where AI output comes from. Those efforts are happening right beside our own work.
CH 06 Two moves close to pharma practice
Things AI produces often cannot be traced later. Who made them, and how?Where and how to keep that origin is the next question.One effort places a mark inside a substance that AI designed. If the mark goes in without harming how the substance works, its origin can be traced. The other is a drug maker tidying its core records before spreading AI across the company. It chose to fix the data underneath before the models on top. Both aim to keep outputs and inputs explainable later.A mark helps only if it leaves the contents working as before. If it weakened the result, nobody would use it. Showing that it need not do so makes origin tracking a more practical choice for users.Pharma has always earned trust through records. If you cannot show who checked what, and when, a submission will not be accepted. That principle holds with AI too. How complete your origin records and core data are will set how much work you can hand to AI. Apart from tidying origins and records, though, some questions still have no answer from anyone.
CH 07 What is being overlooked
Outside the big headlines, several things remain unsettled. Listing what is unknown also shows what today's big stories are missing.One major company head is expected to stay away when lawmakers question risky AI. Avoiding that while under review could shake trust in the company. A company under scrutiny needs to explain itself in public. Another statute, said to apply from today, was also reported. Which one, and what it says, is still unclear. One developer claimed a rival nation's well-known system can mount strong attacks. Yet here, no outside party has confirmed that claim. Claims about a competitor are safer left out of decisions until evidence appears.Recording the unknown as unknown is also a core skill in practice. Keep verified and unverified reports apart, and later judgments go wrong less often. Companies that note open questions can update internal rules fast once a rule's content is clear. So, with the certain and the uncertain kept apart, what can users actually rely on?
Wrap-up
What users can rely on is not a builder's promise. It is the records they can check for themselves. When outside rules shift, firms that can verify their tools' condition and origin stay calm. Tomorrow, we watch how the firms under review respond. Will they answer with words, or with records?
That difference is a first clue to how serious the promises are.
- CH 01New York Post「Exclusive | FTC opens sweeping probe of Anthropic, OpenAI and other 'super intelligence' models」 nypost.com
- CH 01cnbc.com「FTC is investigating OpenAI, Anthropic and other AI companies over product risks」 cnbc.com
- CH 02TechCrunch「Pledge signed by President Trump and top AI leaders misspells the United States」 techcrunch.com
- CH 02The New York Times「Opinion | Trump Says A.I. Titans Should Police Themselves. What Could Go Wrong?」 nytimes.com
- CH 02Futura, le média qui explore le monde「"10% is not unreasonable": an AI pioneer warns of extinction risk」 futura-sciences.com
- CH 02Nautilus | Science Connected「“Morally Binding” AI Safety Pledge Mocked」 nautil.us
- CH 03The New York Times「DeepSeek and Huawei Target a Key Source of Nvidia’s A.I. Dominance」 nytimes.com
- CH 03Reuters「DeepSeek partners with Huawei to develop chip programming tools, reducing reliance on Nvidia」 reuters.com
- CH 03Seeking Alpha「DeepSeek unveils Huawei AI chip software to challenge Nvidia」 seekingalpha.com
- CH 04The Pharma Letter「Ortet launches with $500 million commitment for healthcare AI」 thepharmaletter.com
- CH 05SEAD: A State-Based Perspective on Attack and Defense in Tool-Using Agents（「実行の前に読み取りだけで状態を調べてから許可・遮断を決める防御を提案した。」）
- CH 05Daily Papers: SEAD（「論文共有の場での支持票は 2 件、コメントは 3 件、実装リポジトリの star は 1 件である。」）
- CH 06Google DeepMind「Introducing SynthID Bio」 deepmind.google
- CH 06Nature「Function-preserving watermarking of AI-generated proteins」 nature.com
- CH 06Morningstar「Servier Adopts Veeva OpenData Globally to Scale AI」 morningstar.com
- CH 07NBC News「OpenAI CEO Sam Altman to skip congressional hearing on rogue AI agents」 nbcnews.com
- CH 07dailycampus.com「New AI regulation law goes into effect Oct. 1」 dailycampus.com
- CH 07Tom's Hardware「Anthropic claims popular Chinese AI model has Mythos-class hacking abilities」 tomshardware.com
Articles used
Today's related reports
- AI Daily News October 1, 2026
The narration is synthetic speech. The content draws only on this site's articles from the same day. Each edition passes seven plain-language checks before publication. That means known obstacles to comprehension are held below threshold — it is not a guarantee of comprehension.