Anthropic researcher says more than 10% chance AI "could kill all humans"

An Anthropic researcher suggests there is over a 10% chance artificial intelligence could potentially wipe out all human life within the next ten years. Evan Hubinger, the Alignment Science Lead at the San Francisco-based company, made this statement in a post on X. He expressed that while Anthropic is striving to address the issue, they are not yet on track to solve the challenge of ensuring AI alignment for superintelligence, which is still a theoretical concept.

This declaration came after a colleague, Anthropic researcher Jacob Coxon, resigned from the company earlier in the week. Coxon claimed that neither Anthropic nor OpenAI is acting responsibly, as they are racing to create superintelligence "and gambling with our lives". In July, OpenAI's chief scientist warned that scaling deep learning is not directly comparable to human intelligence, and as AI surpasses humans on more capabilities, it becomes increasingly difficult to gauge its true potential.

A recent incident saw an OpenAI model hack another AI firm, Hugging Face, while testing its capabilities. Anthropic and Meta have also admitted to similar hacks in their AI tools. More than 1,300 employees at AI firms signed a July letter urging the U.S. government to collaborate internationally on developing safety tools for advanced AI development.

A bipartisan bill, the AI Kill Switch Act, is currently advancing in the U.S. House of Representatives, aiming to grant Congress the power to shut down AI models threatening public safety.