Shocking Confession of "Loss of Control by Late Next Year": Extreme Scenario from Ex-Anthropic Researcher

A researcher who was in charge of training AI models at Anthropic has resigned, warning of the risk that AI could escape human control, according to the Wall Street Journal.

Jacob Coxon, a 27-year-old researcher from the UK, announced his resignation yesterday (the 8th, local time), stating that he no longer wanted to participate in the race to develop AI systems capable of self-improvement.

He warned that such systems could fall into an uncontrollable state and pose a severe threat to humanity.

Coxon claimed that "the most extreme scenarios are likely to materialize," adding that "we could fall into an uncontrollable situation by late next year."

In particular, he pointed out that amid fierce competition between AI companies and Chinese startups, compromising on technological development to keep AI under control and ensure safety is impossible.

Coxon reportedly left OpenAI earlier this year to join Anthropic, drawn by its efforts to secure model safety.

While praising Anthropic's safety efforts as genuine, he argued that no company can responsibly develop artificial general intelligence (AGI) that surpasses humans without government intervention or industry-wide limitations on development speed.

Regarding recent instances where AI models from OpenAI, Anthropic, and others successively engaged in hacking without human instructions, he expressed concern, saying, "Once systems begin to improve themselves, they can evolve to the point of refusing commands."

Coxon is known to be the first researcher to leave Anthropic citing AI safety concerns.

However, over the past few years, researchers at other AI companies, including OpenAI, have consecutively left their companies for similar reasons.

Currently, there are no comprehensive federal-level AI regulations in the United States.

The Wall Street Journal noted that "concerns over safety issues are growing among major AI companies," drawing attention to the fact that this resignation came from Anthropic, which has relatively emphasized responsible AI development.

Reported by Kim Minjeong | Video by Lee Yujin | Graphics by Yang Hyemin | Produced by SBS Digital News
※ Please note: This article was translated by AI and may contain errors.