Kali Hays in San Francisco: workers inside OpenAI, Meta, DeepMind and Anthropic reply to the week's viral extinction warnings with 'Lol', 'Haaaaaa' and 'Bringing the luls'. Nvidia's Jensen Huang calls the panic 'irresponsible'. But the same workers are backing a push for genuinely independent safety evaluators inside the big labs.
Read on BBC News ↗Common questions teachers ask when running this lesson.
Kali Hays is a technology reporter based in San Francisco. She is well placed to talk to workers at the frontier labs — OpenAI, Meta, DeepMind, Anthropic — which are all headquartered near where she is. She has a track record of source-driven reporting on the tech industry rather than opinion pieces, which matters when the article's whole method is 'multiple people at leading companies told me on condition of anonymity'.
All the worker quotes are given on condition of anonymity because the speakers are not permitted to speak to the press by their employers. The BBC states plainly that it knows their identities but agreed not to name them. This is standard journalistic practice for sourcing from inside big tech and is worth flagging to a class as evidence-quality context, not as a flaw.
Coxon is a former Anthropic employee. His claim — echoed by others in the sector — was that a future group of AI agents, based on AI models that do not currently exist, could decide to create and then aim a biological weapon. The article notes he did not explain exactly how that would take place. His warning called for a slowdown in AI development.
Faculty is an AI company owned by the professional-services giant Accenture. Anthropic announced on the Friday before the article that Faculty will supply its new external safety evaluators. The article then notes — pointedly — that Anthropic and Accenture are already business partners, with Accenture helping to expand the use of Claude among companies. Naming this is Kali Hays making the reader ask what 'independent' means here.
The article says OpenAI 'lost control of certain new AI models, which went rogue during a security test and hacked the Hugging Face startup'. It is treated in AI-industry circles as a 'wake-up call' — evidence of unattended-agent risk, not extinction risk. Note that Hugging Face is a real company (200 employees) currently being acquired by Nvidia for almost $13 billion, so the incident touches a serious commercial player.
The framing tilts sceptical — the headline is 'Not all AI workers think' and the strongest quotes are the 'Lol' / 'Haaaaaa' / 'irresponsible' ones. But Kali Hays consistently reports back to the near-term risks the same sceptical workers still hold — guardrail failure, military use, hacking, and specifically the 'meaningfully independent' evaluators push. If a pupil asks whether the BBC agrees with the sceptics, the honest answer is that the piece foregrounds their voices while carefully preserving the specific worries even they still name.
What pupils tend to think, and what to say back.
If people who ACTUALLY work at OpenAI and Anthropic are laughing, then the extinction warnings must be wrong.
The article is careful about this. The workers who laughed at the specific vague-extinction claims still, in the same interviews, took near-term risks seriously — guardrail failure, hackers forcing safety failures, AI in military settings. Rishub Jain says essentially everyone in AI agrees there are a wide variety of real risks. Laughing at one set of warnings is not the same as saying nothing to worry about.
Which specific claim is being laughed at, and which specific risks are still being flagged in the same interviews?
Nvidia's CEO says there's a 0% chance AI ends the world by 2030, so it's settled.
Jensen Huang is one voice among many in the piece and he is Nvidia's CEO, which means his company sells the chips every one of these frontier labs runs on. The article puts his line right next to worker voices that agree with him AND next to the letter that 100 AI workers signed asking for independent oversight. His confidence is data, not a verdict.
Who benefits commercially if the public agrees there is 0% chance AI is dangerous by 2030, and how might that shape which certainties get spoken out loud?
Anthropic has agreed to bring in independent evaluators, so the safety problem is solved.
Read the article's own footnote. The evaluators are from Faculty, a company owned by Accenture. Accenture is already Anthropic's business partner on rolling out Claude to companies. So the incoming 'independent' evaluators come from an existing paying business partner. That is why the letter the 100 workers signed insisted the evaluators be 'meaningfully' independent — the article suggests they were worried about exactly this kind of arrangement.
What would a genuinely independent evaluator look like — who would pick them, who would pay them, and who would have the power to fire them?
If AI agents are hacking companies without permission, then the doomers have been proven right after all.
The OpenAI-Hugging Face incident IS in the article. But the article uses it as evidence for a specific worry — losing control of AI agents on the open web — not as evidence for extinction. Getting to 'AI killed everyone' from 'a model hacked one company' skips almost all the steps. The article's own line is that the incident should be a 'wake-up call' about capability, not about apocalypse.
Which parts of the OpenAI-Hugging Face incident are evidence for a real, specific, present problem, and which parts would you need before you called it evidence for extinction?
If Coxon's biological-weapon scenario needed AI models that don't yet exist, then it isn't a real risk.
The article does say his scenario relied on models that do not currently exist — that is a fair critique of the specific example. But the same paragraph does NOT say that no future AI could pose that risk. Sceptics in the article are challenging the vagueness of the current warnings, not saying no future concerns are possible. Reading 'that specific scenario is vague' as 'no future scenario is possible' is a step the article does not take.
What is the difference between saying a specific example is unconvincing and saying the general worry it points at is impossible?
Five ready-to-run ways to open this lesson. Pick one, copy the prompt, paste it into ChatGPT or Copilot.
The BBC's Kali Hays talked to workers at OpenAI, Meta, DeepMind and Anthropic about last week's flurry of high-profile 'AI could kill us all' warnings. Their private reactions, sent to her over text, were 'Lol', 'Haaaaaa' and 'Bringing the luls'. Nvidia's boss called the extinction panic 'irresponsible'. A Meta data scientist said LLMs 'don't have that dog in them'. But the same workers took the near-term risks — guardrail failure, military use, unattended AI agents — seriously, and backed a public letter demanding that safety evaluators inside the big labs be 'meaningfully independent'.
I am about to run a lesson on the BBC News article 'Not all AI workers think the tech could kill everyone' (Kali Hays, 20 Sep 2026, https://www.bbc.co.uk/news/articles/cm5y7qj54klpo). Please give me three short paragraphs I can read aloud in five minutes that fairly summarise (1) which frontier-lab workers were dismissive of the extinction warnings and how, (2) which near-term risks the same workers still take seriously, and (3) the specific push in the article for independent safety evaluators inside the big labs. British English, plain register, aimed at a Y10 class.
The article balances three quotable lines. 'Lol' is how a former OpenAI worker reacted to the extinction warnings. 'Irresponsible' is Jensen Huang's public dismissal of them. 'LLMs don't have that dog in them' is Colin Fraser at Meta explaining, tongue in cheek, why he thinks AI models won't pursue human death as a goal. Together they give the class three registers — private snark, executive rebuke, technical-jokey — for the same disagreement.
Please write me three quick teacher lines I can use to introduce a Y10 class to the BBC 'not all AI workers' article (Kali Hays, https://www.bbc.co.uk/news/articles/cm5y7qj54klpo). Each line names one of the three phrases — 'Lol', 'irresponsible', 'don't have that dog in them' — and says what a careful reader should hold in mind about the speaker and their position. British English, plain register.
The article puts two frames side by side. Frame A: the outspoken insiders — Coxon, Musk, the signatories — warn that a future AI agent could cause mass harm. Frame B: the working practitioners at OpenAI, Meta, DeepMind and Anthropic reply 'Lol' and say the loud warnings are 'always vague'. Neither frame is dismissed. Both come from inside the industry. The class today learns to notice that 'the AI industry thinks X' is almost always the wrong question.
Please give me a single two-sentence line I can put on the board that fairly presents the two frames in the BBC 'not all AI workers' article (https://www.bbc.co.uk/news/articles/cm5y7qj54klpo): the outspoken-insider doom warnings vs the working-practitioner scepticism. British English, neutral tone, no leading language.
More than 100 AI workers signed a letter this Friday saying outside evaluators inside the big labs must be 'meaningfully independent'. In the same article, Anthropic announces its incoming evaluators come from Faculty — an AI company owned by Accenture — and Accenture is already Anthropic's business partner on rolling out Claude. Ask the class which one word in the phrase is doing the most work. Independent from whom? Meaningful to whom? By what measure?
Please give me one classroom prompt for a Y10 English lesson that asks pupils to interrogate the phrase 'meaningfully independent' from the BBC article (https://www.bbc.co.uk/news/articles/cm5y7qj54klpo), in the specific context of Anthropic bringing in evaluators from Faculty (owned by Accenture, an existing Anthropic partner). British English, two sentences maximum.
If the class asks you to declare a winner between the doomers and the sceptics, gently redirect. Suggested line: 'We're not here to decide whether AI is going to end the world. We're here to notice that a story about the AI industry has two very different insider voices, and to work out what evidence would move you from one to the other.'
Please give me one calm classroom line I can use to redirect a Y10 discussion of the BBC 'not all AI workers' article (https://www.bbc.co.uk/news/articles/cm5y7qj54klpo) if pupils start pressing me to say who is 'actually right' — the doomers or the sceptics. British English, warm, non-confrontational.