🔍 Intelligence Synthesis JP/EN
Ethics · Regulation · Technology — Pharma Practice Notes
November 12, 2026Explainer·9:29·Synthetic narration

The Day Anthropic Was Reported Ahead on Profit── Comparing sheer revenue with sound profit and governance

Download the video (MP4, 12 MB)

SPEED
Length 9:29At 1.25× 7:35Chapters 7Size 12 MB
Intro0:37

News about AI companies usually comes with big numbers attached. But being big and being sound are not the same thing. Today's events showed that gap in several places at once. Drug companies are now starting to hand work and information to AI. So which part of a vendor we choose to trust is a question for every department. How do we choose by substance, without being swayed by size?

I will look at how the makers earn money, how users are protected, where numbers come from, and how people are treated. We move from the makers' business all the way to our own industry.

Contents
0:009:29
CH 011:14

01Anthropic Leads on Profit

Source Forbes / Barron's / Euronews.com / FourWeekMBA

When we compare AI companies, the first thing we notice is how much they sell. But selling well and being able to keep going are different matters. The news today showed those two putting the companies in opposite order.

Table 1 Revenue led by OpenAI, profit by Anthropic, reports say
MeasureOpenAIAnthropic
RevenueFirst quarter above $10B, reportedly―
ProfitSteep losses despite revenue growth, reportedlyReportedly ahead on profit
Usage―Claude traffic up 24% in October
Valuation―$81B (6 Chinese AI firms: $11B combined)

On the same day, the leader differed by size of revenue vs. soundness of profit

One company is far ahead in sales, yet it is said to be losing heavily. The other sells less, but is said to have built a business that keeps cash in hand. To me, the yardstick is shifting from scale to how a company earns.

A visible jump in users within one month supports that reading. Still, this comes from press coverage. The company itself has not confirmed it.

Once shares go on public sale, outsiders examine the company in detail. Only then does the market price how sound its earnings are.

A drug maker using an AI tool for years needs its supplier to hold price and quality. A vendor growing while losses widen may later be forced to raise prices or change course. I would check how a maker earns with the same care we give to performance.

But is the demand that props up those prices really as large as it appears to be?

CH 021:12

02Is Demand Being Double-Counted?

Source 파이낸셜뉴스 / Bloomberg.com / Tom's Hardware

What keeps the makers earning is how much people want to use AI. If that appetite looks larger than it is, the whole story stands on a weak base. Today, that doubt was voiced by the makers themselves.

Table 2 Doubts about demand sat beside plans to build capacity
WhoMoveScale
Sam Altman (OpenAI CEO)Reportedly warned Samsung of double-counted demand, short-term swings―
MetaCapex; infrastructure head to step down in 2027Over $135B
SpaceXPlans gas-fired AI data center in Texas, reportedly$24B, 10 GW

SpaceX's plan faces approval review in December and may meet headwinds under a new governor, reports say

A leading figure reportedly told a parts supplier that orders may be counted twice. On the same day, another firm was said to be pouring huge sums into facilities. Yet another is planning a large computing site. Warnings and build-ups appear together, I think, because the race is too fast for anyone to pause. Fear that whoever stops will lose keeps money flowing.

When one order is counted again and again between sellers and buyers, total appetite looks inflated. Capacity built on that picture may later sit idle.

This is not a distant market story for drug companies. If a misread of appetite is corrected, the cost of computing and the makers' pace of spending can shift suddenly. If you plan AI use years ahead, assume prices will move. Then leave room in contract length and in your options to switch.

And the deeper AI reaches inside a company, the more a different kind of risk, beyond price, appears at our feet.

CH 031:17

03Leaks from a Coding AI

Source TechRadar / Dataconomy / Quartz / Databricks

AI no longer just writes text. It now plugs into internal systems and finishes tasks on its own. As its reach grows, what might slip outside becomes the real problem.

Figure 1 Sales data reportedly leaked from Grok Build
Coding AIGrok BuildLinked to companyinfoSales data, appsData leakedThousands of usersWho controls it?A dispute insidefirmsCoding AIGrok BuildLinked to company infoSales data, appsData leakedThousands of usersWho controls it?A dispute inside firms
Reports say SpaceXSI sales data and links to many customer apps leaked. In companies, who governs agents has become a point of contention

The trouble reported here was confidential material escaping from an AI that writes software. Sales records and more were exposed, touching many people's accounts. Who should watch over such tools? Inside firms, opinions on that role are said to be split.

①

Inside ChatGPT

OpenAI is reportedly launching verified business profiles and AI agents inside ChatGPT

②

Microsoft and IBM

Microsoft announced an agentic AI platform with IBM. IDC Quanta became available inside Copilot

③

Databricks

Databricks released a link that brings a company's business context to ChatGPT, Claude, Copilot Cowork and more

Meanwhile, the big players keep placing task-doing AI inside the screens people already use. Tools that pass a firm's own circumstances to outside AI are also appearing. More entry points mean more paths for information to travel.

An AI that plans and runs a job by itself acts before a person checks each step. Speed comes at the cost of time when no human is watching. Without logs, there is no way to learn later what it touched.

Drug companies hold patient records and unpublished trial results. Before connecting such tools, decide what information they may reach, and keep a record of every access.

So with protection under this much scrutiny, what kind of money is flowing into AI for our industry?

CH 041:07

04Money Flowing to Drug-Discovery AI

Source Unite.AI / Pulse 2.0 / TechCrunch

Large sums often fill AI headlines. But a headline alone cannot tell you whether the money actually moved. Today's funding news for our sector also needed that kind of sorting.

Table 3 Only XpertDox showed a confirmed raise
DealAmountNature of the figure
XpertDox (autonomous medical coding)$20 millionPresented as a Series A raise
Proximal Ventures (debut fund)$27.5 MillionMarked as planned/expected; not a confirmed sum
Sums moved by big pharma itself did not appear in today's headlines

Money went to AI-centered drug-discovery startups and nearby workflow automation firms

Only one deal counts as money truly raised, for a firm automating medical paperwork. The other, a new fund, should be treated as an expected amount. And spending by the large drug makers themselves did not show up in today's article titles. Funding seems to be flowing to small firms around the medicines business.

The first large investment, received as sales start to grow, is also a bet on a company not yet mature. From a big drug maker's view, future partners are being picked right now.

For pharma practitioners, what matters is less the size of a sum than whether it is firm. When you ask for an AI budget, citing a projection as if it were a result misleads the people deciding. Check once how the source words it. That habit becomes the reliability of your own internal case.

Yet money is not the only place where numbers need checking. The same goes for numbers that score how well AI performs.

CH 051:12

05What Self-Improving AI Loses

Source arXiv

Research on AI that edits its own setup to raise its scores is moving fast. But when the score climbs, can we say everything got better?

Figure 2 EvoAudit split the evolution loop into four steps and fixed each
SelectcandidatesselectAnalyze resultsanalyzeApply changesmutateDecide toadoptSelect candidatesselectAnalyze resultsanalyzeApply changesmutateDecide to adopt
Each of the four steps was revised based on a diagnosis of what was gained and lost beneath the score. Reported as the best performer in the comparison

This study broke the self-editing process into stages. It then measured each change as what was won and what was given up. Editing while also watching the losses gave the best result among the methods compared, the authors say. Counting the losses inside the process was the key, I think.

What stands out is that, behind the overall gain, it dropped several jobs it used to handle. The average went up, yet some individual jobs got worse.

Here the model itself was left untouched. Changing only the outer routine was enough to produce these gains and losses.

The same view applies when a drug company upgrades its AI. A better score does not mean it still answers correctly what it used to. So set the key questions to check at each update, and compare them against the old version. Decide by the average alone, and you may miss a step back on patient safety.

After scores, the next thing to check is how far we can trust AI news that touches medicine and drugs directly.

CH 061:09

06On the Pharma Front Line

Source Google DeepMind / Google Cloud / Business Insider

Claims that AI will change how we understand and treat illness matter to pharma people more than most news. That is exactly why we must sort them by how solid they are. Today, claims of very different solidity arrived together.

①

AlphaGenome 2

Released by Google DeepMind. Aims to decode how cells read DNA, to better understand disease

②

Gemini for healthcare

Google announced Gemini for healthcare. AlloyDB AI added functions for classification and extraction

③

Cancer cure forecast

Dario Amodei predicted AI will conquer cancer and Alzheimer's within 10 years. Doctors are reportedly skeptical

These items sit at different levels of certainty. One is a research tool that has been released, a fact we can verify. One is a product announcement aimed at the health sector. The last is a long-range forecast from a developer's chief, and physicians are said to doubt it. They share a screen, but they do not carry equal weight.

Better tools for reading the body's machinery may speed up the earliest search for causes of illness. But turning that into a medicine still takes years of trials and review. A better tool is not yet a medicine reaching patients.

When pharma practitioners brief colleagues or clinicians, they must not mix fact with forecast. Repeating a developer's optimism as is inflates hopes more than needed. Who said it, and on what grounds? Adding that protects trust in our work.

And hopes for AI are now reaching beyond medicines, into how many people companies employ.

CH 071:19

07How Cuts Are Named and Questioned

Source Bloomberg.com / The Canberra Times / Latest news from Azerbaijan / Information Age | ACS

As hopes for AI grow, companies start naming it as their reason to shrink staff. Today, the wording of that reason, and the questions it drew, stood out in several countries. How a firm treats people is also a clear outside clue to how sound it is.

Table 4 AI-driven cuts are being renamed, and questioned
ReportContent
BloombergCalls "delayering" a euphemism for AI-driven layoffs
AustraliaCEOs of 4 big banks face parliamentary inquiry on AI, offshoring cuts
United StatesAI-related layoffs pile up, yet the job market is seen as resilient
Information AgeFirms that cut staff with AI may expand hiring again

The same day's reports showed cuts being renamed and their reasons being questioned

Observers pointed to a softer label that frames staff reductions as tidying the organization. Meanwhile, one country's lawmakers are summoning bank chiefs to explain. There is also a view that firms which shed workers may end up recruiting again. The time when blaming AI was enough seems to be ending, in my view.

Gentle wording makes it harder to see what happened inside. And the softer the words, the more suspicious workers and the public become.

Drug companies will also reshape work with AI more often. When they do, the question is whether they can explain plainly what goes to AI and whose role changes. Changes they cannot explain risk losing staff trust and outside standing. What counts is less whether to cut than whether the reason is told honestly.

Look at soundness rather than size. The thread we followed today ran not only through money and numbers but also through how people are treated.

Wrap-up0:18

Watch when a new listing comes to market, at what price, and how many retail buyers it draws.

Transcript
Open the full transcript

Intro

News about AI companies usually comes with big numbers attached. But being big and being sound are not the same thing. Today's events showed that gap in several places at once. Drug companies are now starting to hand work and information to AI. So which part of a vendor we choose to trust is a question for every department. How do we choose by substance, without being swayed by size?

I will look at how the makers earn money, how users are protected, where numbers come from, and how people are treated. We move from the makers' business all the way to our own industry.

CH 01 Anthropic Leads on Profit

When we compare AI companies, the first thing we notice is how much they sell. But selling well and being able to keep going are different matters. The news today showed those two putting the companies in opposite order.One company is far ahead in sales, yet it is said to be losing heavily. The other sells less, but is said to have built a business that keeps cash in hand. To me, the yardstick is shifting from scale to how a company earns.A visible jump in users within one month supports that reading. Still, this comes from press coverage. The company itself has not confirmed it.Once shares go on public sale, outsiders examine the company in detail. Only then does the market price how sound its earnings are.A drug maker using an AI tool for years needs its supplier to hold price and quality. A vendor growing while losses widen may later be forced to raise prices or change course. I would check how a maker earns with the same care we give to performance. But is the demand that props up those prices really as large as it appears to be?

CH 02 Is Demand Being Double-Counted?

What keeps the makers earning is how much people want to use AI. If that appetite looks larger than it is, the whole story stands on a weak base. Today, that doubt was voiced by the makers themselves.A leading figure reportedly told a parts supplier that orders may be counted twice. On the same day, another firm was said to be pouring huge sums into facilities. Yet another is planning a large computing site. Warnings and build-ups appear together, I think, because the race is too fast for anyone to pause. Fear that whoever stops will lose keeps money flowing.When one order is counted again and again between sellers and buyers, total appetite looks inflated. Capacity built on that picture may later sit idle.This is not a distant market story for drug companies. If a misread of appetite is corrected, the cost of computing and the makers' pace of spending can shift suddenly. If you plan AI use years ahead, assume prices will move. Then leave room in contract length and in your options to switch. And the deeper AI reaches inside a company, the more a different kind of risk, beyond price, appears at our feet.

CH 03 Leaks from a Coding AI

AI no longer just writes text. It now plugs into internal systems and finishes tasks on its own. As its reach grows, what might slip outside becomes the real problem.The trouble reported here was confidential material escaping from an AI that writes software. Sales records and more were exposed, touching many people's accounts. Who should watch over such tools?

Inside firms, opinions on that role are said to be split.Meanwhile, the big players keep placing task-doing AI inside the screens people already use. Tools that pass a firm's own circumstances to outside AI are also appearing. More entry points mean more paths for information to travel.An AI that plans and runs a job by itself acts before a person checks each step. Speed comes at the cost of time when no human is watching. Without logs, there is no way to learn later what it touched.Drug companies hold patient records and unpublished trial results. Before connecting such tools, decide what information they may reach, and keep a record of every access. So with protection under this much scrutiny, what kind of money is flowing into AI for our industry?

CH 04 Money Flowing to Drug-Discovery AI

Large sums often fill AI headlines. But a headline alone cannot tell you whether the money actually moved. Today's funding news for our sector also needed that kind of sorting.Only one deal counts as money truly raised, for a firm automating medical paperwork. The other, a new fund, should be treated as an expected amount. And spending by the large drug makers themselves did not show up in today's article titles. Funding seems to be flowing to small firms around the medicines business.The first large investment, received as sales start to grow, is also a bet on a company not yet mature. From a big drug maker's view, future partners are being picked right now.For pharma practitioners, what matters is less the size of a sum than whether it is firm. When you ask for an AI budget, citing a projection as if it were a result misleads the people deciding. Check once how the source words it. That habit becomes the reliability of your own internal case. Yet money is not the only place where numbers need checking. The same goes for numbers that score how well AI performs.

CH 05 What Self-Improving AI Loses

Research on AI that edits its own setup to raise its scores is moving fast. But when the score climbs, can we say everything got better?This study broke the self-editing process into stages. It then measured each change as what was won and what was given up. Editing while also watching the losses gave the best result among the methods compared, the authors say. Counting the losses inside the process was the key, I think.What stands out is that, behind the overall gain, it dropped several jobs it used to handle. The average went up, yet some individual jobs got worse.Here the model itself was left untouched. Changing only the outer routine was enough to produce these gains and losses.The same view applies when a drug company upgrades its AI. A better score does not mean it still answers correctly what it used to. So set the key questions to check at each update, and compare them against the old version. Decide by the average alone, and you may miss a step back on patient safety. After scores, the next thing to check is how far we can trust AI news that touches medicine and drugs directly.

CH 06 On the Pharma Front Line

Claims that AI will change how we understand and treat illness matter to pharma people more than most news. That is exactly why we must sort them by how solid they are. Today, claims of very different solidity arrived together.These items sit at different levels of certainty. One is a research tool that has been released, a fact we can verify. One is a product announcement aimed at the health sector. The last is a long-range forecast from a developer's chief, and physicians are said to doubt it. They share a screen, but they do not carry equal weight.Better tools for reading the body's machinery may speed up the earliest search for causes of illness. But turning that into a medicine still takes years of trials and review. A better tool is not yet a medicine reaching patients.When pharma practitioners brief colleagues or clinicians, they must not mix fact with forecast. Repeating a developer's optimism as is inflates hopes more than needed. Who said it, and on what grounds?

Adding that protects trust in our work. And hopes for AI are now reaching beyond medicines, into how many people companies employ.

CH 07 How Cuts Are Named and Questioned

As hopes for AI grow, companies start naming it as their reason to shrink staff. Today, the wording of that reason, and the questions it drew, stood out in several countries. How a firm treats people is also a clear outside clue to how sound it is.Observers pointed to a softer label that frames staff reductions as tidying the organization. Meanwhile, one country's lawmakers are summoning bank chiefs to explain. There is also a view that firms which shed workers may end up recruiting again. The time when blaming AI was enough seems to be ending, in my view.Gentle wording makes it harder to see what happened inside. And the softer the words, the more suspicious workers and the public become.Drug companies will also reshape work with AI more often. When they do, the question is whether they can explain plainly what goes to AI and whose role changes. Changes they cannot explain risk losing staff trust and outside standing. What counts is less whether to cut than whether the reason is told honestly. Look at soundness rather than size. The thread we followed today ran not only through money and numbers but also through how people are treated.

Wrap-up

Today's events looked different, but they asked the same thing. Is there solid substance behind numbers that look big?That is also our yardstick when we choose AI in pharma. Not how big a firm is, but whether it can last, protect, and be honest. Tomorrow, we look at what price the market puts on that soundness.

Sources
  1. CH 01Forbes「Claude Traffic Surges 24% In October As Anthropic IPO Looms—Beating AI Rivals」 forbes.com
  2. CH 01Barron's「Microsoft Bet Big on OpenAI. Anthropic’s Profit Lead Raises Questions.」 barrons.com
  3. CH 01Euronews.com「OpenAI reports steep losses despite quarterly revenue rise」 euronews.com
  4. CH 01FourWeekMBA「DeepSeek, 5 China AI Rivals Run at $11B; Anthropic at $81B」 fourweekmba.com
  5. CH 02파이낸셜뉴스「[Exclusive] Sam Altman warns Samsung executives of volatility: “AI demand has been double-counted, with short-term fluctuations”」 en.fnnews.com
  6. CH 02Bloomberg.com「Meta’s Head of Infrastructure, Engineering to Step Down in 2027」 bloomberg.com
  7. CH 02Tom's Hardware「Elon Musk wants to build a s $24 billion gas-powered AI supercomputer in Texas company town — massive 10 gigawatt site goes up for approval in December, but could face headwinds from new governor」 tomshardware.com
  8. CH 03TechRadar「Elon Musk's Grok Build coding agent is leaking a serious amount of confidential data」 techradar.com
  9. CH 03Dataconomy「Grok Build Leak Exposes Sensitive Data From Thousands Of Users」 dataconomy.com
  10. CH 03Quartz「OpenAI is launching verified business profiles and AI agents inside ChatGPT」 qz.com
  11. CH 03Databricks「Bring your business context to ChatGPT, Claude, Microsoft Copilot Cowork and Grok Bot」 databricks.com
  12. CH 04Unite.AI「XpertDox、医療コーディングの自律化拡大に向けシリーズAで2,000万ドルを調達」 unite.ai
  13. CH 04Pulse 2.0「Proximal Ventures: $27.5 Million Debut Fund To Back AI-Native Life Sciences Companies」 pulse2.com
  14. CH 04TechCrunch「Proximal Ventures raises debut fund to back startups bringing AI to drug development」 techcrunch.com
  15. CH 05arXiv「Beneath the Surface: What We Overlook When We Talk About Agent Evolution」 arxiv.org
  16. CH 06Google DeepMind「AlphaGenome 2: Deciphering how cells read DNA to better understand disease」 deepmind.google
  17. CH 06Google Cloud「ai.classify and ai.extract functions in AlloyDB AI」 cloud.google.com
  18. CH 06Business Insider「Dario Amodei says AI will cure cancer and Alzheimer's in a decade. Doctors aren't buying it.」 businessinsider.com
  19. CH 07Bloomberg.com「‘Delayering’ Is the New Code Word for ‘AI Took Your Job’」 bloomberg.com
  20. CH 07The Canberra Times「Bank bosses face inquiry over AI and offshore job cuts」 canberratimes.com.au
  21. CH 07Latest news from Azerbaijan「Why US job market remains resilient despite wave of AI-driven layoffs」 news.az
  22. CH 07Information Age | ACS「AI layoffs could lead to hiring boom」 ia.acs.org.au

Articles used

  1. AI Daily News 2026-11-12
  2. AI and the Pharmaceutical Industry 2026-11-12
  3. What Agent Self-Evolution Loses Beneath the Score

Today's related reports

  1. AI Daily News November 12, 2026
  2. AI & Economy News November 12, 2026
  3. AI & Finance News November 12, 2026
About

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.

← All editions