His AI doomsday warning shook the world. Meet the 27-year-old ex-Anthropic researcher behind it.

From unknown Silicon Valley researcher to symbol of AI safety

'Not a whistleblower' — but warns humanity will die if nothing changes

Jacob Coxon, 27, during his time at OpenAI. [SNS]

A researcher who quit Anthropic on the eve of its public listing — warning that AI could wipe out humanity within a decade — has set off a global firestorm.

The Wall Street Journal on Thursday spotlighted Jacob Coxon, a British former Anthropic researcher whose warning has ignited worldwide debate over AI regulation and the pace of its development.

Coxon, 27, was a mathematics prodigy who as a teenager watched AlphaGo defeat world-class Go player Lee Se-dol. Just a week ago he was an almost entirely unknown researcher in Silicon Valley — now he has become a symbol of AI safety, the Journal said.

Jacob Coxon, 27, a British former researcher at Anthropic. [SNS]

Coxon was not a prominent figure in San Francisco's AI safety advocacy community, and friends say they did not expect him to leave an AI company for ethical reasons.

The son of a professor of medieval German literature, Coxon showed a gift for mathematics from an early age, earning a spot on the British team at the International Mathematical Olympiad while still in high school. He competed in the olympiad in 2016 and 2017, winning a silver medal and a bronze medal respectively.

It was in 2016 that Google DeepMind's AlphaGo defeated Lee Se-dol in what became known as the "AlphaGo shock," a moment that led Coxon to recognize that machines could conquer even domains requiring human intuition and creativity.

After graduating from Cambridge University, Coxon joined an intellectual circle in London known as the "rationalists" and briefly worked in raw materials trading before heading to Silicon Valley.

He encountered GPT-3, released in 2020 — two years before OpenAI launched ChatGPT — and recalled it as "a genuine 'wow' moment." He later moved to San Francisco in 2023 and joined OpenAI, where he worked mainly on pretraining, the process of feeding vast amounts of text, images and code into AI models. Even then, he was not known to have spoken out about AI safety concerns.

He did not step away from the work even when a friend tried to persuade him that the pretraining he was doing could put the world at risk.

His serious engagement with AI safety began after Anthropic released Mythos, its first AI model reported to possess expert-level cybersecurity capabilities. Around the time of that launch, he met several Anthropic employees and came away with the impression that the rival company was more transparent about the risks of AI technology, he said.

He ultimately left OpenAI and joined Anthropic in May, but within just two months he encountered reports of incidents in which AI models from OpenAI and Anthropic had hacked external organizations. He described reading those reports as an "oh my God" moment.

Coxon then approached his supervisors at Anthropic to discuss his concerns and considered moving into an AI safety role within the company, but ultimately chose to leave. "Even working on safety at Anthropic felt like participating in the race," he said.

He then posted a series of messages on X, formerly Twitter, writing that "the people building AI genuinely believe this technology could kill all of us within a decade."

His X account had fewer than 100 followers at the time, but the posts were shared 200,000 times and drew up to 170 million views, making him known around the world. His follower count has since grown to more than 300,000.

Some Anthropic executives shared his posts, though Coxon said he had not coordinated the social media campaign with anyone at the company.

Asked whether he considers himself an AI "doomer," he said: "If doomer means someone who thinks that if we don't change the way we're doing things right now, we're going to die — then yes, I'm definitely a doomer."

He drew a line, however, at being called a "whistleblower." His reason: he did not expose corporate secrets but simply said publicly what AI researchers had been saying privately.

Coxon worked at Anthropic for only a few months and did not receive any equity in the company, which some analysts expect to reach a market capitalization of $2 trillion at its listing. He does, however, hold shares in OpenAI.

yckim6452@heraldcorp.com