Key Takeaways
- Anthropic researcher resigns, warning that the AI race risks catastrophe.
- Says fear is private, even when public tone sounds calm.
- Joins 1,000+ researchers urging global AI limits.
A researcher who worked on AI training at both OpenAI and Anthropic has resigned, warning that AI labs are moving too fast toward systems that could threaten human life within years.
A Researcher’s Public Warning
Jacob Coxon specialized in pre-training, the process of teaching AI models by feeding them massive amounts of data. He spent three years working on this at two of the industry’s leading AI labs before stepping down from his role.
Coxon shared his reasons for resigning in a public post on social media, explaining why the Anthropic researcher resigns and saying neither company he worked for was acting responsibly. He said both are racing toward building self-improving AI systems without fully accounting for the risks such systems could pose.
He described the technology as one that could soon match or exceed human capabilities across many fields. Coxon said such systems could potentially breach digital security systems, transform entire industries overnight, and gain significant real-world power and resources on their own.
He pointed to steady progress across multiple areas of AI development as evidence that these capabilities are arriving faster than many expect. According to Coxon, this pace shows no signs of slowing down in the near future.
Private Fears Versus Public Statements
According to Coxon, the Anthropic researcher resigns as many people building advanced AI systems privately worry about catastrophic outcomes, even when their public statements sound more measured and reassuring. He said this concern is shared quietly among researchers and executives across the field, regardless of how confident they appear publicly.
Coxon suggested a divide exists between the two companies where he worked. He said many OpenAI staff have not fully internalized the scale of the risks involved in their work. At Anthropic, he said staff understand the stakes clearly but continue development anyway. Their reasoning, he explained, is a belief that competitors will not act responsibly, so they must move forward first despite the danger.
He also noted that colleagues across the industry increasingly use terms like “crunchtime” and “endgame” to describe how close the field may be to building self-improving systems capable of advancing without human guidance.
Call for Slower Development
Coxon urged fellow AI researchers to reconsider the pace of their work and think carefully about what the coming years could bring. He raised pointed questions about whether launching advanced training systems without a rigorous understanding of their behavior was a responsible path forward.
He asked researchers to consider whether continuing simply because “it’s happening anyway” was the right approach or whether this moment called for pushing back and demanding different conditions across the industry.
Coxon was among more than 1,000 AI researchers who recently signed a statement calling for coordinated global action on AI safety. The Anthropic researcher resigns statement calls for mechanisms that could slow AI development if systems begin improving on their own in ways that become difficult to control or predict.
His departure marks one of the first known instances in which an Anthropic researcher resigns from a major AI lab specifically over safety concerns tied to the pace of development, according to reporting on the resignation. His comments add to a growing conversation within the AI research community about balancing rapid technological progress with long-term safety planning.
Neither Anthropic nor OpenAI has issued a detailed public response addressing Coxon’s specific claims about internal attitudes toward AI risk within their organizations.
Visit more of our news! Future Education Magazine