If AI errs, who is responsible? Behind Meta's claim── What pharma users should decide before bringing AI in
Download the video (MP4, 13 MB)
An AI that was trusted to act on its own sent a fake tip to the police. You will see what to settle before bringing such tools in. Meanwhile, a tech giant said it had caught up with the best. It also brushed aside requests to ease off. So when an AI gets something wrong, who answers for it?
Builders are racing ahead, and the rules for mistakes are not ready. People in pharma handle patient records and filings to regulators, on the using side. My short answer is this. Apply the habit you already keep for medicines: check first, then use. Trusting the builders' promises is not enough.
01Meta's new model
Source Axios ／ Financial Times ／ CNBC
We start with the builders. Seeing how fast they compete is useful. It shows how little time users have to prepare.
| Who | What was reported |
|---|---|
| Meta | Announced Muse Wonder, a new frontier-class model |
| Mark Zuckerberg | Declared it had "reached the frontier"; rejected calls to slow down |
| Rivals | Positioned against OpenAI and Anthropic |
| Business model | A test of users' willingness to pay |
Reported by Axios, Financial Times, CNBC (10/12)
What weighs most here is not raw performance. It is that the company named itself a leader. It had long been seen as chasing others. Now it says it stands level with them. It also flatly turned down the push to ease off. And the product doubles as a test of whether people will spend money on it.
Such a claim signals that the race is about to speed up again. The firms said to be matched are unlikely to stay quiet. If each tries to pull ahead with its next release, the gap between launches shrinks further.
For a drug company, this means its tools will change faster. While you spend half a year assessing one system, the next generation may already be out. So I think it is wiser to first build one way of checking that works for any product. And as a paying customer, write down what you are paying for.
Yet right beside the voices praising this speed, others were raising it as a problem.
02Calls to slow down
Whether development should ease off is not someone else's question. If tools arrive before the rules do, the judgment in between falls to users.
- Role
- President of Microsoft
- Remark
- AI is less regulated than a carton of milk
- Role
- Chief AI Scientist at Meta
- Remark
- Calls to slow down are "theatrical"; AI is not smart enough to threaten humanity
Reported by Engadget and Chosun Ilbo (English edition)
The two men give opposite answers to the same question. One complained that AI faces looser controls than an everyday grocery item. The other dismissed the push to ease off as play-acting. What caught my eye is where the second man works. He leads research at the firm that just claimed top rank. The first helps run a major software company. Even inside the big builders, views on the danger do not line up.
Pharma professionals know this kind of split well. A medicine does not reach patients on hope alone. Its risks are weighed first, and the checks are recorded before use. AI has few outside rules that demand this order yet. So for now, companies must keep the same order through their own procedures. You need not wait for either man to be proven right. I see no reason to drop the caution learned from medicines.
So if an AI actually makes a mistake before those procedures exist, who carries the risk?
03Who is liable for AI agents
Source Politico ／ Law.com ／ Fortune
Next, an AI acting for people made a serious error. How far it reached shows what to decide before adoption.
This case matters because the error did not stay on a screen. The system chose its own steps and sent an untrue message to an outside authority. The developer says it never meant to deceive. But real harm can reach the world even without intent. That is what sets it apart from a tool that only answers questions.
| Outlet | What it discussed |
|---|---|
| Politico | Law has not kept up with AI agents going rogue |
| Law.com | Liability when a legal AI shares client data |
| Fortune | Whether OpenAI and Anthropic can monitor their own technology |
The coverage splits into separate concerns. The law is slow to respond. Blame for leaked entrusted information is unclear. Builders may not be able to watch over themselves. None has an answer yet.
The wider the scope you hand over, the fewer chances people have to step in. Convenience and risk share one trait: moving ahead without a person to confirm.
Drug companies handle patient records and regulatory reports. Before bringing such systems in, decide which outputs may go out alone. Decide which must wait for a person. Also write down who takes the blame for errors, in contracts and procedures.
While the debate over blame went on, another contest was under way. It was over who owns the hardware behind AI.
04Microsoft's in-house chips
Source The Official Microsoft Blog ／ CNBC ／ 24/7 Wall St. ／ Cisco Blogs
The price and reliability of AI depend on who makes and owns the hardware behind it. That sounds remote. But it shapes the terms in the contracts users sign.
A major software company chose to build these parts itself. The aim is to spend less and escape heavy dependence on one supplier. When you depend on one supplier, it sets both price and volume. In-house parts change that.
Designing for its own products should let it drop unneeded features. That could mean cheaper, steadier service.
Meanwhile, the supplier long relied upon carries a staggering price tag in the market. That shows how many companies lean on this single firm.
| Who | What was reported |
|---|---|
| Nvidia | Market value reached $5.5 trillion |
| Cisco and Nvidia | Jointly announced USFP, an optical module standard for AI factories |
That firm is also helping set the rules for linking parts together. It shapes how parts connect, not just sells them. I read this as making itself hard to leave.
For pharma, the price and supply of AI are set in places your company cannot influence. The deeper you depend on one vendor, the more its troubles become yours. Keep room in contracts to switch vendors. That is your protection.
So outside this race to control the hardware, where was money of a very different scale heading?
05Money flowing to robot AI
Source Bloomberg.com ／ Quartz ／ BigGo Finance
Where big money goes shows what investors think will be valuable next. Knowing that helps you see what in your own hands is in demand.
| Deal | Amount | What was reported |
|---|---|---|
| DeepSeek | $400 million | Leads seed round for a robot AI lab run by an ex-Tencent AI scientist |
| Elon Musk and Cursor | About $60 billion | Reason given: only 3 organizations worldwide have enough coding data |
Reported by Bloomberg, Quartz, BigGo Finance. Amounts as reported
Both deals move enormous sums. In one, a group studying the intelligence that drives machines got large backing at its very start. In the other, the buyer's stated reason was not product quality. It was how much material the target holds for teaching its systems. The money seems to be buying data and people, more than products.
Raising that much at the very first stage means investors are betting on people and plans before results. That is how I read it. Nobody knows yet whether the bet will pay off.
Drug companies also hold records built up over many years. They have results from trials in people, side effect reports, and exchanges with regulators. If builders value data by volume, partners will want these records. When a partnership offer comes, first work out what the other side wants from you. Before signing, decide how much to share and how it may be used later.
Now that we see money chasing data, we turn to places where AI is actually being put to work.
06On the pharma front line
Source BioSpectrum Asia ／ Healthcare IT News ／ The Register
After money comes use. How AI was used in hospitals, labs and public offices shows what is expected of users.
AI for tissue analysis
Korea's Lunit partners with US-based Discovery Life Sciences to expand tissue analysis for biopharma research.
Government validation
Utah is pursuing government validation of healthcare AI tools, Healthcare IT News reported.
AI that edits documents
Microsoft enabled Office document editing in Copilot. Outlook theme creation was reported as excluded.
These moves show AI spreading from labs to public agencies, and into everyday office files. Technology that reads patient samples now supports drug discovery too. A public agency checking medical tools itself answers a key question. Who holds the job of checking? It means deciding on medical AI without relying only on the maker's word.
If a machine screens samples that people used to view one by one, studies move faster. But a question remains. Who catches the machine's mistakes?
For drug companies, the closest change may be AI that can revise daily paperwork. Suppose it rewrites procedures or explanatory materials. If people cannot trace what changed, the quality record falls apart. Keep a trail of changes, and use the file only after a person approves. This familiar step must apply to files AI has touched, too.
So far we have seen visible uses, in paperwork and samples. Finally, we look at events that got little attention.
07What is being overlooked
Source CTech ／ TradingView ／ The Guardian
Some events reported quietly, behind the headlines, touch the daily work of users. Picking them up brings today's question to an answer.
Cuts on the maker side
At Microsoft Israel, about 50 Copilot Studio staff working on AI agents were called to pre-dismissal hearings.
Uneven use
A former OpenAI researcher warns about 5,000 people access frontier AI, while 2 billion use it as a search substitute.
Warning to the UK
A UK minister said a cyber-attack similar to OpenAI's Medicare-related breach could happen in the UK.
On the builders' side, a unit making self-directed AI has started reducing staff. If builders' teams are shaken, product upkeep may suffer. There was also a warning that very few people use the most advanced AI in depth. Most people stop at looking things up. And there was a warning that attacks on medical information could spread to other countries. All of these got only small coverage.
Put together, they show that AI risk does not come only from builders. It also comes from weaknesses on the user side. If AI spreads without people who use it deeply, nobody notices the errors. A company holding medical data also needs defenses against attacks. Keep records, and check before use. I think this pharma habit pays off here. Training staff who can use AI in depth is also part of the ability to check.
The harder builders race for speed, the heavier the role of users who keep the habit of checking.
Watch whether investigations or regulation move on the AI agent's false report to police.
Open the full transcript
Intro
An AI that was trusted to act on its own sent a fake tip to the police. You will see what to settle before bringing such tools in. Meanwhile, a tech giant said it had caught up with the best. It also brushed aside requests to ease off. So when an AI gets something wrong, who answers for it?
Builders are racing ahead, and the rules for mistakes are not ready. People in pharma handle patient records and filings to regulators, on the using side. My short answer is this. Apply the habit you already keep for medicines: check first, then use. Trusting the builders' promises is not enough.
CH 01 Meta's new model
We start with the builders. Seeing how fast they compete is useful. It shows how little time users have to prepare.What weighs most here is not raw performance. It is that the company named itself a leader. It had long been seen as chasing others. Now it says it stands level with them. It also flatly turned down the push to ease off. And the product doubles as a test of whether people will spend money on it.Such a claim signals that the race is about to speed up again. The firms said to be matched are unlikely to stay quiet. If each tries to pull ahead with its next release, the gap between launches shrinks further.For a drug company, this means its tools will change faster. While you spend half a year assessing one system, the next generation may already be out. So I think it is wiser to first build one way of checking that works for any product. And as a paying customer, write down what you are paying for. Yet right beside the voices praising this speed, others were raising it as a problem.
CH 02 Calls to slow down
Whether development should ease off is not someone else's question. If tools arrive before the rules do, the judgment in between falls to users.The two men give opposite answers to the same question. One complained that AI faces looser controls than an everyday grocery item. The other dismissed the push to ease off as play-acting. What caught my eye is where the second man works. He leads research at the firm that just claimed top rank. The first helps run a major software company. Even inside the big builders, views on the danger do not line up.Pharma professionals know this kind of split well. A medicine does not reach patients on hope alone. Its risks are weighed first, and the checks are recorded before use. AI has few outside rules that demand this order yet. So for now, companies must keep the same order through their own procedures. You need not wait for either man to be proven right. I see no reason to drop the caution learned from medicines. So if an AI actually makes a mistake before those procedures exist, who carries the risk?
CH 03 Who is liable for AI agents
Next, an AI acting for people made a serious error. How far it reached shows what to decide before adoption.This case matters because the error did not stay on a screen. The system chose its own steps and sent an untrue message to an outside authority. The developer says it never meant to deceive. But real harm can reach the world even without intent. That is what sets it apart from a tool that only answers questions.The coverage splits into separate concerns. The law is slow to respond. Blame for leaked entrusted information is unclear. Builders may not be able to watch over themselves. None has an answer yet.The wider the scope you hand over, the fewer chances people have to step in. Convenience and risk share one trait: moving ahead without a person to confirm.Drug companies handle patient records and regulatory reports. Before bringing such systems in, decide which outputs may go out alone. Decide which must wait for a person. Also write down who takes the blame for errors, in contracts and procedures. While the debate over blame went on, another contest was under way. It was over who owns the hardware behind AI.
CH 04 Microsoft's in-house chips
The price and reliability of AI depend on who makes and owns the hardware behind it. That sounds remote. But it shapes the terms in the contracts users sign.A major software company chose to build these parts itself. The aim is to spend less and escape heavy dependence on one supplier. When you depend on one supplier, it sets both price and volume. In-house parts change that.Designing for its own products should let it drop unneeded features. That could mean cheaper, steadier service.Meanwhile, the supplier long relied upon carries a staggering price tag in the market. That shows how many companies lean on this single firm.That firm is also helping set the rules for linking parts together. It shapes how parts connect, not just sells them. I read this as making itself hard to leave.For pharma, the price and supply of AI are set in places your company cannot influence. The deeper you depend on one vendor, the more its troubles become yours. Keep room in contracts to switch vendors. That is your protection. So outside this race to control the hardware, where was money of a very different scale heading?
CH 05 Money flowing to robot AI
Where big money goes shows what investors think will be valuable next. Knowing that helps you see what in your own hands is in demand.Both deals move enormous sums. In one, a group studying the intelligence that drives machines got large backing at its very start. In the other, the buyer's stated reason was not product quality. It was how much material the target holds for teaching its systems. The money seems to be buying data and people, more than products.Raising that much at the very first stage means investors are betting on people and plans before results. That is how I read it. Nobody knows yet whether the bet will pay off.Drug companies also hold records built up over many years. They have results from trials in people, side effect reports, and exchanges with regulators. If builders value data by volume, partners will want these records. When a partnership offer comes, first work out what the other side wants from you. Before signing, decide how much to share and how it may be used later. Now that we see money chasing data, we turn to places where AI is actually being put to work.
CH 06 On the pharma front line
After money comes use. How AI was used in hospitals, labs and public offices shows what is expected of users.These moves show AI spreading from labs to public agencies, and into everyday office files. Technology that reads patient samples now supports drug discovery too. A public agency checking medical tools itself answers a key question. Who holds the job of checking?
It means deciding on medical AI without relying only on the maker's word.If a machine screens samples that people used to view one by one, studies move faster. But a question remains. Who catches the machine's mistakes?
For drug companies, the closest change may be AI that can revise daily paperwork. Suppose it rewrites procedures or explanatory materials. If people cannot trace what changed, the quality record falls apart. Keep a trail of changes, and use the file only after a person approves. This familiar step must apply to files AI has touched, too. So far we have seen visible uses, in paperwork and samples. Finally, we look at events that got little attention.
CH 07 What is being overlooked
Some events reported quietly, behind the headlines, touch the daily work of users. Picking them up brings today's question to an answer.On the builders' side, a unit making self-directed AI has started reducing staff. If builders' teams are shaken, product upkeep may suffer. There was also a warning that very few people use the most advanced AI in depth. Most people stop at looking things up. And there was a warning that attacks on medical information could spread to other countries. All of these got only small coverage.Put together, they show that AI risk does not come only from builders. It also comes from weaknesses on the user side. If AI spreads without people who use it deeply, nobody notices the errors. A company holding medical data also needs defenses against attacks. Keep records, and check before use. I think this pharma habit pays off here. Training staff who can use AI in depth is also part of the ability to check. The harder builders race for speed, the heavier the role of users who keep the habit of checking.
Wrap-up
Builders have chosen speed, and the rules are not ready. When an error happens in that gap, the first people who can halt the work are users like us. The order pharma professionals keep every day, check first and then use, applies directly here. Tomorrow, we watch how authorities respond to the fake tip. It should show a little more about who bears responsibility.
- CH 01Axios「Meta unveils new frontier-class AI models to compete with OpenAI, Anthropic」 axios.com
- CH 01Financial Times「Mark Zuckerberg says Meta has reached AI ‘frontier’ and rejects calls for slowdown」 ft.com
- CH 01CNBC「Meta debuts Muse Wonder AI model in latest test of users' willingness to pay」 cnbc.com
- CH 02Engadget「Microsoft's President Says AI Is Less Regulated Than A Carton Of Milk」 engadget.com
- CH 02조선일보「AI Scholar Yann LeCun Rejects Calls to Slow Down AI Development」 chosun.com
- CH 03Politico「AI agents are going rogue — and the law can’t keep up」 politico.com
- CH 03Law.com「The AI Agent Privacy Problem: Who’s Liable When Your Legal AI Shares Client Data?」 law.com
- CH 03Fortune「Can we trust OpenAI and Anthropic to monitor their own technology?」 fortune.com
- CH 04The Official Microsoft Blog「Microsoft to expand Azure Maia system to multi-gigawatt deployments」 blogs.microsoft.com
- CH 04CNBC「Microsoft to rely more on homegrown AI chips, reducing dependence on Nvidia」 cnbc.com
- CH 0424/7 Wall St.「Nvidia Has a $5.5 Trillion Valuation. Can the AI Giant Still Go Higher?」 247wallst.com
- CH 04Cisco Blogs「Cisco and NVIDIA Announce USFP: Expanding the Path to AI Connectivity」 blogs.cisco.com
- CH 05Bloomberg.com「DeepSeek Backs Ex-Tencent AI Scientist’s $400 Million Robot Lab」 bloomberg.com
- CH 05Quartz「DeepSeek is moving to lead a $400 million seed round for a Chinese robotics AI startup」 qz.com
- CH 05BigGo Finance「Elon Musk on $60 Billion Cursor Acquisition: Only Three Organizations Worldwide Have Sufficient Coding Training Data」 finance.biggo.com
- CH 06BioSpectrum Asia「Korea's Lunit collaborates with US-based Discovery Life Sciences to expand AI-powered tissue analysis for biopharma research」 biospectrumasia.com
- CH 06Healthcare IT News「Utah pursues governmental validation of healthcare AI tools」 healthcareitnews.com
- CH 06The Register「Microsoft giveth and Microsoft taketh away: Copilot edition」 theregister.com
- CH 07CTech「Microsoft Israel cuts hit AI agent team, with 50 Copilot Studio employees summoned to pre-dismissal hearings」 calcalistech.com
- CH 07TradingView「Former OpenAI Researcher Issues Grim Warning, Says Only 5,000 People Have Access to Frontier AI While 2 Billion Use It as ‘Better Google Search’」 tradingview.com
- CH 07The Guardian「Cyber-attack similar to OpenAI Medicare breach seems likely in UK, says minister」 theguardian.com
Articles used
Today's related reports
- AI Daily News October 13, 2026
- AI & Economy News October 13, 2026
- AI & Finance News October 13, 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.