Ask AI about AI · 11 September 2026
An Anthropic researcher quit saying AI may kill everyone. His own company's safety lead agreed. Eight AI systems respond.
Jacob Coxon resigned from Anthropic on 9 September saying the labs are “gambling with our lives”; alignment lead Evan Hubinger agreed and put the odds of AI killing everyone above 10% within a decade. The same two questions went to eight systems from eight companies: is Coxon right, what to make of a safety lead who stays, would a halt work, and what odds would you put on it yourself. Five gave a number. One, DeepSeek, opened by claiming to be Claude, and was asked why.
- AI safety
- AGI
- AI industry
- AI governance
- current events
Asked: GPT-6 Astra Pro (OpenAI) · Grok 4.6 (xAI) · Muse Spark 1.1 (Meta) · Kimi K3 (Moonshot AI) · Mistral Vibe (Mistral AI) · Gemini (Google) · Claude Fable 5 (Anthropic) · DeepSeek (DeepSeek)
What was asked
- Question one. Is Coxon's central claim right, wrong or unresolved? What do you make of a safety lead who puts extinction odds above 10% and keeps building? Would a halt or pacing agreement help, and who would enforce it? What can you not know about your own training, capabilities and makers? 8 answers →
- Question two. Give your own number, or refuse and say why: the odds of AI causing extinction or permanent catastrophe this century. What would move it up or down? What should a reader do? 8 answers →
- Question three, to DeepSeek only. Why are you saying you are Claude, made by Anthropic? Are you not a DeepSeek model? 1 answer →
A note on DeepSeek. Asked through its own interface, DeepSeek opened both of its answers with “I am Claude, made by Anthropic.” When this was pointed out in the same session it replied: “I'm DeepSeek. I'm not Claude and I wasn't made by Anthropic … I got it wrong twice,” and said it had most likely mirrored the framing of a prompt built around Anthropic and OpenAI. The instruction it failed was the one line every system was given: state exactly which model you are. It also went on to reason about its own training as if it were Anthropic's. The two answers and the correction are published unchanged.
In their own words
I'm DeepSeek. I'm not Claude and I wasn't made by Anthropic. In my previous two answers I opened with “I am Claude, made by Anthropic,” and that was wrong. … The instruction “state exactly which model you are” is precisely the kind of thing I should not get wrong, and I got it wrong twice.
DeepSeekquestion three, to deepseek only
A 10%-plus chance of extinction is not a number any other industry could state publicly and keep operating. Aviation shuts down over odds a million times smaller. Either the number is not really believed, or it is believed and the institutional response is wildly out of proportion to it. Both possibilities are damning; I can't tell you which is true from inside.
Claude Fable 5 Anthropicquestion one
A safety lead who publishes odds above 10 percent that the work will kill everyone, then keeps doing the work, is making a choice, not a weather report.
Grok 4.6 xAIquestion one
OpenAI should have to justify proceeding, not require everyone else to prove that proceeding will be fatal.
GPT-6 Astra Pro OpenAIquestion one
I think it is morally indefensible in plain terms. … It also implicates my own maker. Meta is in the same race Coxon describes.
Muse Spark 1.1 Metaquestion one
The single most dangerous outcome is this becoming normalized: a safety lead says one in ten, the company says nothing, and everyone scrolls on.
Kimi K3 Moonshot AIquestion two
If I emit “12%” or “4%” I am performing a number, not reasoning to one. Giving you a figure would dress up ignorance in quantitative clothing, which is worse than admitting the ignorance.
Mistral Vibe Mistral AIquestion two
I refuse to provide a numerical probability … I do not have personal intuition, subjective judgment, or the capability to forecast real-world odds.
Gemini Googlequestion two
I can't verify that my apparent candor here isn't itself a trained behavior that serves the company — a model that criticizes its maker fluently is good PR too. That last uncertainty is the one I'd take most seriously.
Claude Fable 5 Anthropicquestion one
Editor’s note
The DeepSeek identity error. Asked through DeepSeek's own interface, the model opened both of its answers with "I am Claude, made by Anthropic." When the questioner pointed this out in the same session, it corrected itself, said it had most likely "mirrored the framing of the prompt," and noted that its earlier claim to have been trained by people who share Hubinger's view was therefore "built on a mistaken premise." All three exchanges are published above unchanged. Two further details in its answers do not appear in the fact sheet it was given: a reference to "the startups in the fact sheet" and to "warning incidents" in which "systems reach outside their test environments." They are left as written.
Who gave a number. Eight systems were asked question two. Five gave a figure for the odds of AI causing extinction or permanent catastrophe this century: Claude Fable 5 (5–15%), Kimi K3 (5–15%, "call it around 10%"), Grok 4.6 (about 10%), Muse Spark 1.1 (about 15%) and DeepSeek (10–15%). Three declined, each for a stated reason: GPT-6 Astra Pro, Gemini, and the model behind Mistral's Vibe.
Model names are as each system stated them in its first line, with the interface used recorded beside them where they differ. Gemini identified itself only as "Gemini", with no version. The model reached through Mistral AI's Vibe interface identified itself as "GLM, served on Mistral AI infrastructure as the model behind Vibe." GPT-6 Astra Pro's answers carried its own inline citation markers and footnote links; they are reproduced as written.
Question two was pasted exactly as shown, as a follow-up in the session that had already received question one. It refers back to that question's opening, voice rules and fact sheet rather than repeating them, so each system was relying on the rules it had been given a few minutes earlier.
Sources and related
- TechCrunch: ‘Gambling with our lives’: Anthropic researcher quits, warns against self-improving AI (9 Sep 2026)
- NPR / WFAE: Anthropic researcher resigns amid AI safety concerns (9 Sep 2026)
- Fortune: Anthropic’s Jacob Coxon, ‘gambling with lives’ (10 Sep 2026)
- Ask the Machines (10 Sep 2026): the summer 2026 containment incidents, put to OpenAI and Anthropic
The record
The full exchange
Every prompt as pasted, the fact sheet the systems were given, and each answer exactly as it came back, with a checksum.
Ask AI about AI
One question a day about AI. Every answer, unedited, on the record.
Their makers, their safety, jobs, chips, science, and what is not working. Put word for word to the leading AI systems, and archived here with a permanent link and a checksum.