The Day OpenAI Pledged to Halt NDAs── Post-election stock slide, AI costs and jobs, money for AI drug discovery
Download the video (MP4, 12 MB)
AI is often discussed in terms of how smart it can get. But today's news suggests something else decides whether it moves forward. It is the trust of the people around it, more than raw performance. Neighbors of the sites where computing is built are judging it. So are investors who hold shares, employees at work, and scientists making medicines. Each group is starting to decide how far AI can be relied on. For anyone hoping to expand AI in pharma, this matters directly. Questioned on money, information and people, would your own plan earn that trust?
01OpenAI's Pledge to Local Residents
Source OpenAI ／ CNBC ／ Reuters ／ Politico
Our first scene is the day after the vote. Builders of computing sites had just faced anger at the polls. What did they offer the towns in return?
| Item | What was reported |
|---|---|
| Nondisclosure agreements | Pledged to halt them under its data center policy |
| Local residents | Listen to their views on construction |
| Construction plans | Make the plans public |
| Electricity bills | Protect residents who pay the bills |
CNBC reported the policy followed the election rebuke. Politico called it a new policy framework
They offered openness, not new technology. Plans once kept quiet will now be shared with neighbors. Households will be shielded from the cost of power. Put another way, secrecy no longer gets a site built. Local consent now comes before raw computing power.
Why the secrecy until now? Early word of a site's location or size can push up land prices or spark opposition. That is a reasonable guess, but the reports do not say. Silence gave builders speed. It left neighbors distrustful. The vote showed how heavy that distrust had become.
Drug companies face a similar pattern. Delay explaining a new plant or lab to the community, and the same backlash can follow. Rolling out AI inside a company works the same way. Tell the people involved what it will do and what goes into it, before it starts. That order decides how fast adoption moves.
After the neighbors, the market took its own measure of trust in AI. Where did the money go once the votes were counted?
02AI Stocks Fall After the Election
Source Investing.com ／ Yahoo Finance ／ The Times ／ Forbes
Markets change their mood faster than towns do. The morning after the results, where did the money riding on AI head?
| Who | What was reported | Outlet |
|---|---|---|
| Elon Musk | Lost billions as AI and data center shares fell | The Times |
| Jensen Huang | Fell to No. 9 on billionaire list amid chip selloff | Forbes |
| Anthropic | AI liquidation behind US stock plunge clouds its mega-IPO | Investing.com |
| Michael Burry | Concerns over updated S-1; won't buy IPO shares | Yahoo Finance |
It moved away. Shares of the companies that power AI were sold. Founders saw their fortunes shrink sharply. A developer getting ready to list was said to feel the chill. A well-known investor read its filing and chose to stay out. Prices built on endless growth are now being questioned.
Why does the timing of a listing matter so much? For a developer, going public is how it raises cash for computing bills. A cold market means less money comes in.
A filing must also lay out a company's weaknesses. That investor judged its spending promises too heavy. Only extraordinary growth could pay for them, in that reading.
For people in pharma, this is not a distant stock story. The tools you use today run on these companies' funding. If that money thins, prices may rise or features may shrink. It pays to plan, at the contract stage, for not depending on one vendor.
When money tightens, companies often turn to people first. Next, we look at the workforce.
03AI Costs and Voluntary Buyouts
Source BigGo ファイナンス ／ Yahoo!ニュース ／ NNAアジア ／ SMH.com.au
Where does the huge money for computing come from? Part of the answer shows up in staff numbers.
The logic is simple. The more a company pours into computing, the harder it must squeeze other spending. So it offers exit terms and lets employees choose to go. Yet inside the same company, the teams building AI are growing. I read this as replacing people more than reducing them.
| Company | Scale | Reported by |
|---|---|---|
| Microsoft | Buyouts offered to about 60,000, about 27% of all staff | Yahoo! News Japan |
| Canva | Restructuring covering 22% of all staff | NNA Asia |
| Heidi | Cut 100+ jobs weeks after a large funding round | SMH.com.au |
Other firms have recently announced exits or reorganizations reaching a big share of their workforce. Even a young company fresh from raising money is letting people go. Profits and cash no longer protect headcount. Spending on AI is changing the assumptions behind hiring.
For a company, letting staff pick departure is a way to shrink without conflict. Japan is among the places covered, so workplaces there are not untouched.
Drug companies may face the same choice one day. So decide early which work goes to AI and which stays with people. Chase savings alone without that line, and your most knowledgeable people may leave first.
Is some of the money saved flowing into AI for medicines? Next, we follow the funding for drug research.
04Money Flowing to AI Drug Discovery
Source Bloomberg.com ／ www.trendingtopics.eu ／ BioSpectrum India
Even in a cold market, money is still flowing to AI for medicines. What question does that put to drugmakers?
| Company | Amount | Nature of the amount |
|---|---|---|
| Isomorphic Labs | ¥789 billion ($5 billion) | In talks to raise; valuation of $50 billion |
| Venomics AI | $2.7 million | Reportedly raised; venom-based drug discovery |
Isomorphic Labs is an Alphabet unit. It reportedly has no drug in clinical trials yet
On one side is a specialist backed by a giant parent. On the other is a small firm focused on one kind of material. Their sums differ by several orders of magnitude. Within these reports, funding is gathering at the two extremes.
Note that the larger sum is not settled. A figure under negotiation may never happen. And that company's medicines have reportedly not yet been tested in people. The money is paying for expectation, not results.
A company's value on paper works the same way. It reflects hopes for future medicines, not today's sales.
For a drugmaker, this is about choosing partners. Team up with a partner priced on hope, and the terms tilt its way. Decide how much you will pay, and for results at which stage. Having that yardstick in-house supports every partnership decision.
After money for hope comes information. Once it goes into a model, can it ever be taken back?
05Erasing Data a Model Has Learned
Source IDSISA: Influence-Distilled Sharded Training for Efficient Machine Unlearning
What happens if someone asks you to erase information an AI has already absorbed? Anyone handling patient records must face it.
Today's study answers it by changing how the model is built. Information likely to need erasing is grouped in one place from the start. When a request comes, only that part is relearned. That takes far less effort than rebuilding everything.
Targeted forgetting
Reported better targeted forgetting than SISA. The size of the improvement varied widely by setting.
Performance retention
Reported up to 50% relative improvement in retaining performance versus SISA. The distillation part is approximate.
Time required
Reported that unlearning time rose in some cases and fell in others, depending on the architecture.
The results were mixed. Forgetting the chosen information and keeping accuracy both beat the original method, the authors say. But the gains swung with test conditions, and time sometimes went up. It has not yet passed expert review.
Splitting the learning costs some accuracy. That loss is won back by copying the answers of another model. This part is close, but not a strict guarantee. Only the relearned portion is truly erased.
In pharma, the question is how to respond when a patient withdraws consent. Saying it can't be removed once inside will not be accepted. Decide where information will be learned before it goes in. That design is what backs a promise to protect it.
Only when information is protected does AI become a research tool. Next, we return to where medicines are made.
06In the Pharma Industry
Source Business Wire
Will drugmakers grow their own AI for finding medicines? Or will they hand it to outside specialists?
| Item | What was reported |
|---|---|
| The two companies | Roche and Chai Discovery |
| Field | AI-driven drug discovery |
| Financial terms | Not disclosed |
Announced by Chai Discovery. Details of the scope were not disclosed either
That day, one major company chose to partner with a firm whose core business is AI. Building the same skill in-house takes people and years. Borrowing it from someone who already has it looks faster. Bringing in outside capability has become a visible trend in our industry.
The earliest stage of finding new medicines means testing huge numbers of options, and it is slow. The wish to speed that up is why companies borrow outside strength.
But this time, neither the payment nor the scope has been made public. From outside, we cannot yet judge how serious it is. Look past the announcement to its substance. Which steps are handed over? Who checks the results? How do you exit if it fails? See whether the contract answers these. Only then does a partnership work inside the company.
The more outside help you use, the more vendors you pay. How visible are those costs to the people using the tools?
07Hidden Caps and Costs
Source The Register ／ CRN ／ CryptoSlate
The wider a tool spreads, the bigger the bill. Yet there are surprisingly few clues to how fast it will grow.
You find the cap by hitting it
Copilot Chat has usage caps for users without a paid add-on, and the caps are reported to be hard to see.
Usage billing delayed again
Microsoft again delayed making usage-based billing the default in Copilot Business, citing partner feedback.
Compute needed to verify
Anthropic's FlashProve verifies that computation follows the committed weights. It took 70–95x more GPU hours.
There is a ceiling on use, but you learn its height only when you run into it. The vendor has postponed the switch to paying per use. One report says even the tool that checks whether AI kept its promise is expensive. Each case shows spending staying out of the buyer's view.
Paying for what you use sounds fair. But when consumption is hard to predict, so is the monthly bill. The vendor's delay may reflect buyers' confusion.
This is easy to miss for anyone expanding AI in a drug company. The cost of a pilot does not predict the cost of a company-wide rollout. Estimate before launch where you will hit limits and what checking will cost. Being surprised when the invoice arrives is too late.
Trust lasts only when spending is visible in advance. Now, let's bring today's story to one answer.
Watch whether the Democratic-led Congress puts forward AI and data center regulation.
Open the full transcript
Intro
AI is often discussed in terms of how smart it can get. But today's news suggests something else decides whether it moves forward. It is the trust of the people around it, more than raw performance. Neighbors of the sites where computing is built are judging it. So are investors who hold shares, employees at work, and scientists making medicines. Each group is starting to decide how far AI can be relied on. For anyone hoping to expand AI in pharma, this matters directly. Questioned on money, information and people, would your own plan earn that trust?
CH 01 OpenAI's Pledge to Local Residents
Our first scene is the day after the vote. Builders of computing sites had just faced anger at the polls. What did they offer the towns in return?They offered openness, not new technology. Plans once kept quiet will now be shared with neighbors. Households will be shielded from the cost of power. Put another way, secrecy no longer gets a site built. Local consent now comes before raw computing power.Why the secrecy until now?
Early word of a site's location or size can push up land prices or spark opposition. That is a reasonable guess, but the reports do not say. Silence gave builders speed. It left neighbors distrustful. The vote showed how heavy that distrust had become.Drug companies face a similar pattern. Delay explaining a new plant or lab to the community, and the same backlash can follow. Rolling out AI inside a company works the same way. Tell the people involved what it will do and what goes into it, before it starts. That order decides how fast adoption moves. After the neighbors, the market took its own measure of trust in AI. Where did the money go once the votes were counted?
CH 02 AI Stocks Fall After the Election
Markets change their mood faster than towns do. The morning after the results, where did the money riding on AI head?It moved away. Shares of the companies that power AI were sold. Founders saw their fortunes shrink sharply. A developer getting ready to list was said to feel the chill. A well-known investor read its filing and chose to stay out. Prices built on endless growth are now being questioned.Why does the timing of a listing matter so much?
For a developer, going public is how it raises cash for computing bills. A cold market means less money comes in.A filing must also lay out a company's weaknesses. That investor judged its spending promises too heavy. Only extraordinary growth could pay for them, in that reading.For people in pharma, this is not a distant stock story. The tools you use today run on these companies' funding. If that money thins, prices may rise or features may shrink. It pays to plan, at the contract stage, for not depending on one vendor. When money tightens, companies often turn to people first. Next, we look at the workforce.
CH 03 AI Costs and Voluntary Buyouts
Where does the huge money for computing come from?Part of the answer shows up in staff numbers.The logic is simple. The more a company pours into computing, the harder it must squeeze other spending. So it offers exit terms and lets employees choose to go. Yet inside the same company, the teams building AI are growing. I read this as replacing people more than reducing them.Other firms have recently announced exits or reorganizations reaching a big share of their workforce. Even a young company fresh from raising money is letting people go. Profits and cash no longer protect headcount. Spending on AI is changing the assumptions behind hiring.For a company, letting staff pick departure is a way to shrink without conflict. Japan is among the places covered, so workplaces there are not untouched.Drug companies may face the same choice one day. So decide early which work goes to AI and which stays with people. Chase savings alone without that line, and your most knowledgeable people may leave first. Is some of the money saved flowing into AI for medicines?
Next, we follow the funding for drug research.
CH 04 Money Flowing to AI Drug Discovery
Even in a cold market, money is still flowing to AI for medicines. What question does that put to drugmakers?On one side is a specialist backed by a giant parent. On the other is a small firm focused on one kind of material. Their sums differ by several orders of magnitude. Within these reports, funding is gathering at the two extremes.Note that the larger sum is not settled. A figure under negotiation may never happen. And that company's medicines have reportedly not yet been tested in people. The money is paying for expectation, not results.A company's value on paper works the same way. It reflects hopes for future medicines, not today's sales.For a drugmaker, this is about choosing partners. Team up with a partner priced on hope, and the terms tilt its way. Decide how much you will pay, and for results at which stage. Having that yardstick in-house supports every partnership decision. After money for hope comes information. Once it goes into a model, can it ever be taken back?
CH 05 Erasing Data a Model Has Learned
What happens if someone asks you to erase information an AI has already absorbed?Anyone handling patient records must face it.Today's study answers it by changing how the model is built. Information likely to need erasing is grouped in one place from the start. When a request comes, only that part is relearned. That takes far less effort than rebuilding everything.The results were mixed. Forgetting the chosen information and keeping accuracy both beat the original method, the authors say. But the gains swung with test conditions, and time sometimes went up. It has not yet passed expert review.Splitting the learning costs some accuracy. That loss is won back by copying the answers of another model. This part is close, but not a strict guarantee. Only the relearned portion is truly erased.In pharma, the question is how to respond when a patient withdraws consent. Saying it can't be removed once inside will not be accepted. Decide where information will be learned before it goes in. That design is what backs a promise to protect it. Only when information is protected does AI become a research tool. Next, we return to where medicines are made.
CH 06 In the Pharma Industry
Will drugmakers grow their own AI for finding medicines?Or will they hand it to outside specialists?That day, one major company chose to partner with a firm whose core business is AI. Building the same skill in-house takes people and years. Borrowing it from someone who already has it looks faster. Bringing in outside capability has become a visible trend in our industry.The earliest stage of finding new medicines means testing huge numbers of options, and it is slow. The wish to speed that up is why companies borrow outside strength.But this time, neither the payment nor the scope has been made public. From outside, we cannot yet judge how serious it is. Look past the announcement to its substance. Which steps are handed over?
Who checks the results?How do you exit if it fails?See whether the contract answers these. Only then does a partnership work inside the company. The more outside help you use, the more vendors you pay. How visible are those costs to the people using the tools?
CH 07 Hidden Caps and Costs
The wider a tool spreads, the bigger the bill. Yet there are surprisingly few clues to how fast it will grow.There is a ceiling on use, but you learn its height only when you run into it. The vendor has postponed the switch to paying per use. One report says even the tool that checks whether AI kept its promise is expensive. Each case shows spending staying out of the buyer's view.Paying for what you use sounds fair. But when consumption is hard to predict, so is the monthly bill. The vendor's delay may reflect buyers' confusion.This is easy to miss for anyone expanding AI in a drug company. The cost of a pilot does not predict the cost of a company-wide rollout. Estimate before launch where you will hit limits and what checking will cost. Being surprised when the invoice arrives is too late. Trust lasts only when spending is visible in advance. Now, let's bring today's story to one answer.
Wrap-up
Taken together, today's events show that performance alone does not set AI's path. Neighbors, investors and employees now measure it by their own standards. Those of us using AI in pharma are being measured too. Only a plan that explains the use, the cost, and whose information it holds will earn trust and move ahead. Tomorrow, we watch whether rules for the builders start to move.
- CH 01OpenAI「Setting a new standard for responsible AI infrastructure」 openai.com
- CH 01CNBC「OpenAI pledges to halt NDAs in new data center policy proposal after election rebuke」 cnbc.com
- CH 01Reuters「OpenAI kicks off charm offensive to change voters' view of data centers」 reuters.com
- CH 01Politico「OpenAI lays out new policy framework for data centers」 politico.com
- CH 02Investing.com「Behind Wall Street’s plunge: AI liquidation threatens Anthropic’s mega-IPO」 investing.com
- CH 02Yahoo Finance「Anthropic Is ‘Betting Its Entire Existence On Optimism’: Michael Burry Flags Risks After AI Firm’s Updated S-1, Says Won’t Pursue IPO Shares」 finance.yahoo.com
- CH 02The Times「Elon Musk loses billions as AI and data centre shares fall after midterms」 thetimes.com
- CH 02Forbes「Nvidia’s Jensen Huang Drops To No. 9 In Billionaire Rankings Amid Chipmaker Selloff」 forbes.com
- CH 03BigGo ファイナンス「グーグル、5カ国で希望退職プログラムを拡大 AI・クラウド部門は逆に増員」 finance.biggo.jp
- CH 03Yahoo!ニュース「AIの影響か？ 利益急増でも6万人に退職打診 マイクロソフトの「黒字リストラ」は日本にも広がるか？ #エキスパートトピ」 news.yahoo.co.jp
- CH 03NNAアジア「キャンバが大規模リストラ、ＡＩ競争で成長鈍化」 nna.jp
- CH 03SMH.com.au「AI start-up Heidi cuts jobs weeks after raising $475 million」 smh.com.au
- CH 04Bloomberg.com「アルファベット傘下のAI創薬アイソモーフィック、7890億円調達で協議」 bloomberg.com
- CH 04www.trendingtopics.eu「Isomorphic Labs in Talks for 5 Billion Dollars at 50 Billion Dollar Valuation, Bloomberg Reports」 trendingtopics.eu
- CH 04BioSpectrum India「Venomics AI raises $2.7 M to advance venom-based drug discovery」 biospectrumindia.com
- CH 05IDSISA: Influence-Distilled Sharded Training for Efficient Machine Unlearning（「消去の保証は SISA から引き継がれ、厳密なままである」）
- CH 06Business Wire「Chai Discovery Announces Collaboration with Roche in AI-Driven Drug Discovery」 businesswire.com
- CH 07The Register「Copilot Chat has usage caps, and Microsoft would rather you found them by hitting them」 theregister.com
- CH 07CRN「Microsoft Delays Default Copilot Usage-Based Billing Again, Citing Partner Feedback」 crn.com
- CH 07CryptoSlate「Are you getting the AI you paid for? Anthropic’s new AI verification tool has a costly catch」 cryptoslate.com
Articles used
- AI Daily News 2026-11-06
- AI and the Pharmaceutical Industry 2026-11-06
- AI Daily News headline translations 2026-11-06 morning
- AI Daily News headline translations 2026-11-05 evening
- Reliably erasing specific data from trained models: an improved sharded training method
Today's related reports
- AI Daily News November 6, 2026
- AI & Economy News November 6, 2026
- AI & Finance News November 6, 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.