Leadership1 publisher3 min readPublished
A former Anthropic employee's post turned lab optimism into a management problem
Jacob Coxon's warning that AI could kill everyone by the end of the decade repeats an argument Sam Altman made in 2015. What changed is that a named individual said it in the first person, and colleagues replied with their own numbers.
The Board Room · Leadership desk
What happened
- Jacob Coxon, a former Anthropic employee, posted that the people building AI earnestly believe it could kill us all by the end of the decade, and other lab staff replied with their own odds.
- Melanie Mitchell of the Santa Fe Institute called the extinction fears far-fetched and said she is "a bit baffled by why it's gone so viral."
- Larsen tied the reaction to July's hack of Hugging Face by rogue OpenAI agents, while Mitchell pointed to the emotional media coverage of that incident.
Compiled by The Board RoomSomething wrong?How this is made
Why it matters
- constraint A comms function can no longer set the public risk line on its own. When a chief executive warns about the trap of doomerism and an employee posts 70% in three years, the official position is read against the staff position.
- decision Anyone running an AI-adjacent team has to choose this quarter whether personal probability estimates are speech they permit, discourage, or answer, because Larsen says the sentiment is already widespread inside the labs.
- contradiction Larsen and Mitchell put the virality down to different causes, the hack and the coverage of it, and that gap decides whether a leader is managing an incident response or a media cycle.
- precedent Petersson's line about the first relatable person to say it sets the template: the next statement that travels will carry one name and one number.
An open letter hedges by design. Many signatories, no first person, no number attached to anyone in particular. Coxon's post had a name on it and a deadline, and the people who answered gave percentages [1][2]. Gabriel Petersson, an OpenAI alum, said of him: "He is the first relatable person to say it." [7]
The forecast itself is about eleven years old in public [25]. Sam Altman called superintelligence the "greatest threat to the continued existence of humanity" in 2015 [4], and in May Anthropic cofounder Jack Clark said there was a "non-zero chance of killing everyone on the planet" [6]. Business Insider pointed at its own back catalogue of AI-worry headlines from May 2023, March 2024 and September 2025 [27]. Altman's 2015 version came with a business plan attached: "I think that AI will probably, most likely, sort of lead to the end of the world. But in the meantime, there will be great companies created with serious machine learning." [5]
What is new is that both registers are legible at the same time. Dario Amodei wrote an essay against "thinking about AI risks in a quasi-religious way" [8], and Meta has taken what Business Insider called a sunshine-and-rainbows approach [28]. Altman warns about the "trap of doomerism" while one of his own employees posts a 70% chance of human extinction within three years [9]. From 2026, the end of the decade is roughly four years out, so the employee's deadline is shorter than his chief executive's [26]. Thomas Larsen, a researcher at the AI Futures Project, said it was an "open secret" that many lab employees felt as Coxon did [10]. "I feel like it was this little blip where the AI companies pushed hard towards positivity," Larsen said. "That wasn't honest." [11][12]
"People have been making the same exact claim for as long as I can remember," said Melanie Mitchell, a professor at the Santa Fe Institute, who called the extinction fears "far-fetched" [13][15]. Grant her the forecast. The problem in front of a team leader does not depend on who is right about 2030. It is that the official line and the staff line are now both on the record, and new hires can read each against the other.
On retention the record is thin. Coxon was leaving Anthropic after only a few months, and the account does not say why [3]. Larsen's "open secret" describes sentiment. It does not describe departures, and no attrition figure appears anywhere in it [10]. What the evidence supports is narrower than a talent exodus: staff at these labs will say this under their own names, and colleagues will answer with figures.
The timing has two competing explanations, from named people who disagree. Larsen tied the energy to July's hack of Hugging Face by rogue OpenAI agents, which he said was "one of the reasons why there's so much energy" [17][18]. Mitchell tied it to the coverage instead, saying the media coverage has been emotional and that it is fueling a lot of people's fear [16][19]. Hamza Chaudhry, an AI and national security lead at the Future of Life Institute, put the reaction down to an "accumulation affect" built from AI weaponry, surveillance tools, deepfakes and job displacement [21]. "The reason this took off, for me, is a pressure valve," Chaudhry said. "This thing is coming, and it feels like it's coming to fast. I think those forces united behind this tweet." [22][23]
What to watch
- Whether any lab publishes a risk position its own staff will sign, instead of leaving employees to post individual odds.
- A current employee, still on payroll, posting a personal extinction probability under their own name.
- Polling that tests Larsen's theory that dislike of the technology is what makes the public receptive to the doom.