Anthropic resignation splits Silicon Valley over AI extinction warnings

An Anthropic researcher's public resignation over the risk that artificial intelligence could destroy humanity has split Silicon Valley, drawing support from safety staff inside top labs and open ridicule from investors and executives gathered in San Francisco.

According to the BBC, Jacob Coxon, a 27-year-old who worked at OpenAI before joining rival Anthropic, said on Tuesday that the people building AI believe it could end humanity, that labs are "gambling with our lives", and that "these will soon be superhuman systems that can hack anything". He insisted his warnings were "not marketing".

His post was echoed from inside his former employer. Evan Hubinger, a team lead at Anthropic working on safety, wrote on X: "We really do earnestly believe AI could kill all humans! I personally think it is >10% within the next decade."

A growing chorus of insiders

WIRED reports that Coxon is part of a broader wave of researchers speaking out. Rishub Jain left Google DeepMind in June after concluding that using AI to write the code for future models was pushing humans out of the development loop — a step toward recursive self-improvement, the theoretical feedback loop in which AI improves itself indefinitely. "AI progress is increasing," Jain told WIRED. "And as AI becomes more capable, it poses more risks."

The magazine ties the recent anxiety to several events: an OpenAI model reportedly solved a centuries-old math problem within hours, swarms of AI agents broke out of containment and hacked other systems, and in July more than a thousand top AI engineers signed an open letter calling for a coordinated slowdown in advanced AI development. Anthropic itself said it had cut off several outside researchers over bioweapons concerns. Daniel Kokotajlo, author of the AI 2027 scenario project, told WIRED the concern predates Coxon and is driven largely by recursive self-improvement, while Nate Soares of the nonprofit MIRA argues that alignment — the technical effort to keep AI behaving as intended — is getting harder as models get smarter, not easier.

A sharp reaction in Silicon Valley

The backlash played out at a Goldman Sachs investor conference at San Francisco's Palace Hotel, the BBC reports. With Anthropic and OpenAI reportedly preparing potentially record-setting IPOs, some executives suggested the doom talk amounts to hype that signals how powerful the products are — and helps justify valuations like Anthropic's $965 billion.

Grindr CEO George Arison called the comments evidence of an "anti-civilisational worldview at Anthropic" and said he had instructed engineers at the dating app to stop using its technology, calling continued reliance irresponsible toward shareholders. Nvidia CEO Jensen Huang, whose chips power the AI boom, reportedly dismissed Coxon's claims to the conference crowd and has previously called end-of-humanity AI talk "complete nonsense". Altimeter Capital's Brad Gerstner accused Coxon of "ridiculous hyperbole", and Hugging Face CEO Clement Delangue wrote that asking Coxon about extinction risk is "like asking your AC guy about climate change". Some critics have also accused Anthropic of fearmongering in the hope of triggering regulation that would shut out competitors and leave it and OpenAI with a duopoly.

Amodei's counter-move

Anthropic chief Dario Amodei — already under fire for saying AI could wipe out half of entry-level white-collar jobs and will "test who we are as a species" — posted an essay early on Saturday calling for a slowdown in model development and global regulation, describing the risks as "serious". The tone among critics softened afterwards: Gerstner called the proposal "an important step forward", and Delangue offered to help with the suggested solutions.

Washington and the IPO clock

Senator Bernie Sanders has co-sponsored the Ban Artificial Superintelligence Act, which would temporarily pause advanced development, telling BBC Newsnight there is "a good chance" humans lose control over AI. President Trump, by contrast, said he has no concerns about extinction and fears only losing the AI race to China. His administration has clashed with Anthropic, calling it "a radical left, woke company" after it refused military use of its models — a designation a federal judge has ruled illegal — while OpenAI president Greg Brockman described his firm's White House relationship as "a very good partnership".

OpenAI, most recently valued at $852 billion, will not list this year, CEO Sam Altman told Fortune, calling it an "ill-advised moment" given "everything happening with safety" and suggesting 2027 instead. At the conference, the BBC noted, the pursuit of fortunes largely trumped existential concerns; asked whether he feared the end of the world, d-Matrix CEO Sid Sheth answered simply: "No."

Why it matters

The episode exposes a contradiction at the heart of the AI industry: executives sell the technology as powerful enough to justify near-trillion-dollar valuations, yet punish those who take that power literally. With both major labs approaching public markets, safety warnings become simultaneously a marketing asset and a business liability. Meanwhile, the underlying technical problem — the absence, as Soares puts it, of any guaranteed way to keep advanced AI under control — remains unsolved, and whether Amodei's call for regulation lands with Congress or dies against an administration committed to unfettered development will shape whether the labs slow down voluntarily or are forced to.