September 9, 2026 | 15:11

Reading time: 3 minutes

Evan Hubinger, a senior researcher at Anthropic, warned on X that artificial intelligence could kill all human beings within the next decade. He wrote that he personally considers the probability to be greater than 10% and that, while he believes Anthropic is doing its best, the company does not yet have a plan to solve alignment for a potential superintelligence and is not clearly on track to do so. Hubinger said he regards the risk from current models as low but is worried about a self-improving superintelligence emerging faster than expected; his comment followed a post by a colleague who had just resigned.

In his resignation post, Jacob Coxon wrote that he had left Anthropic after three years working on pre-training research at both OpenAI and Anthropic, and argued that neither company is acting responsibly. He said they are rushing toward a self-improving superintelligence and in doing so are putting lives at risk. According to Coxon, many at OpenAI do not fully grasp the stakes for civilization; at Anthropic the stakes are understood but the company feels trapped in a race to be first, assuming others will not act responsibly and therefore feeling they cannot either, despite the risks.

Jacob is correct here—we really do earnestly believe AI could kill all humans. I personally think it is greater than 10% within the next decade. I believe Anthropic is trying its best, but we do not yet have a plan to solve alignment for superintelligence and are not clearly on track to. https://t.co/QAIHiFP3QZ

— Evan Hubinger (@EvanHub) September 9, 2026

The thread stressed that this technology should not be underestimated. These systems could soon become superhuman in many areas, able to hack systems, transform entire fields rapidly, and acquire real power and resources. Progress has been visible across multiple domains and shows no sign of slowing.

CBS contacted the AI Security Institute, the international organization established by governments to analyze, test and mitigate risks from the most advanced AI systems. A spokesperson from the UK Cabinet Office told CBS that the AI Security Institute continues to work closely with industry partners, including Anthropic, to make models safer, and noted that “just last week” OpenAI’s most powerful model, “GPT-6 Astra,” was tested before public release. The spokesman added that risks cross national borders and no country can handle them alone; the UK will continue testing advanced models, developing a rigorous scientific understanding of their capabilities and risks, and ensuring that policy decisions are grounded in solid evidence.