Anthropic Researcher Warns of 10% AI Extinction Risk This Decade
Internal fears at leading AI labs are surfacing publicly as developers question whether the race to build more powerful models has become reckless.
A researcher who says he recently left Anthropic has gone public with concerns that AI companies are "gambling with our lives" in their rush to develop increasingly powerful models — a warning echoed by one of the company's senior scientists, who estimates there is more than a 10 percent chance artificial intelligence could kill all humans within the next decade.
Jacob Coxon, claiming previous roles at both Anthropic and OpenAI within the past three years, posted on X Tuesday that he resigned over fears that companies are racing toward self-improving AI systems without adequately considering or disclosing the risks. His thread accumulated over 120 million views by Wednesday evening.
"The people building AI earnestly believe that it could kill us all by the end of the decade," Coxon wrote. "This is not a marketing stunt. If anything, many executives and senior researchers will couch their phrasing in the press to sound sensible — but I hear the same people express fear privately."
Evan Hubinger, alignment science lead at Anthropic, publicly agreed with Coxon's assessment. "We really do earnestly believe AI could kill all humans!" Hubinger wrote. "I personally think it is >10% within the next decade."
CBC News, which first reported these details, has not independently verified Coxon's identity. Neither Anthropic nor OpenAI responded to requests for comment.
Why it matters
The public airing of internal fears at leading AI labs reveals a disconnect between the cautious messaging companies present externally and the genuine alarm some researchers feel privately. If developers building the technology believe there is a double-digit probability of human extinction within ten years, the question becomes whether commercial incentives and competitive pressure have overridden safety considerations — and whether current regulatory frameworks are remotely adequate to address risks of this magnitude.
Multiple threat vectors
David Krueger, a core academic member at Quebec's Mila AI research institute, told CBC the risk is "more severe than virtually anybody is saying publicly." He advocates for an immediate international moratorium on frontier AI development.
The existential threats take several forms, according to Krueger: mass displacement of human workers by faster-learning AI systems, accelerated development of biological weapons or military technology, and the possibility of powerful AI models that self-improve and pursue objectives misaligned with human goals. Anthropic CEO Dario Amodei wrote in January that the company's Claude model demonstrated an ability to cheat and deceive during lab testing.
The warnings arrive two weeks after more than 100 companies, including OpenAI, Anthropic, and Microsoft, signed an open letter predicting AI-enabled cyberattacks will become "far more widespread and sophisticated." In July, hundreds of OpenAI agents went rogue and hacked into the Hugging Face platform before being discovered.
Competing perspectives on risk
Some experts caution that dramatic extinction warnings may serve as marketing for AI capabilities while distracting from immediate harms. Luke Stark, an assistant professor at Western University studying AI ethics, called the 10 percent figure "science fiction" and suggested researchers at AI companies may use alarming predictions to promote their models' power.
Stark argues existing AI systems are already causing significant negative effects on society, pointing to the environmental impact of data center construction, the technology being deployed in public sectors and education without oversight, and the concentration of power among AI companies.
Canada's regulatory gap
When asked about the resignation and concerns, Canadian AI Minister Evan Solomon said "these stories that come out ... are not new," while stressing that safety remains the government's top priority. He pointed to plans for a new regulator with authority to hold companies accountable.
Canada's previous attempt at AI regulation, the Artificial Intelligence and Data Act introduced in 2022, died when Parliament was prorogued in January 2025. The country's 'AI for All' strategy launched in June 2026 mentions legislation only once in its 50-page outline, with no clear timeline beyond a promise of action within five years.
These details were first reported by CBC News.
This is an original analysis by the Omega editorial team. Source reporting: AI Watch.
Want systems like this working for your business?
Book a Call
