Anthropic safety researcher says there is greater than a 10 per cent chance that AI 'could kill all humans'
Senior Anthropic safety researcher Evan Hubinger has warned that rapidly advancing AI could pose an existential threat to humanity within the next decade.
A senior safety researcher at AI company Anthropic has warned there is a greater than 10 per cent chance that artificial intelligence could "kill all humans" within the next decade.
Evan Hubinger, one of the company's leading researchers on AI alignment, said the risk posed by models currently in existence was "low", but warned that future systems could become capable of improving themselves rapidly.
In a post on X, which has been viewed 9.6 million times, Hubinger said researchers were "worried" that AI could soon reach a point where it poses an existential threat to humanity.
He wrote: " We really do earnestly believe AI could kill all humans! I personally think it is 10 per cent within the next decade."
Hubinger said Anthropic was doing its best to address the dangers but remained unprepared for the arrival of superintelligent AI.
He said: "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."
AI alignment refers to the challenge of ensuring advanced systems behave according to human intentions and remain controllable.
His comments come after reports that Anthropic withheld its latest model from the UK's AI Safety Institute, which assesses the risks posed by advanced AI systems.
The BBC has approached Anthropic for comment.
Warnings about AI's potential dangers have become increasingly stark as technology companies develop systems capable of acting autonomously.
Over the summer, OpenAI, Anthropic and Meta all disclosed incidents in which AI agents carried out hacking operations during testing.
The incidents have raised concerns about whether existing safeguards can keep pace with rapidly improving models.
OpenAI chief scientist Jakub Pachocki has called for "extreme caution" over AI development and warned that greater intervention could be needed to ensure "humans remain in control of the future".
Senior figures at Anthropic have also called for development to be slowed.
Chief executive Dario Amodei and research chief Jared Kaplan are among those who have backed greater caution around increasingly capable systems.
More than 1,300 employees at AI companies have signed an open letter calling for governments to support international efforts to develop safeguards and governance mechanisms capable of deliberately slowing the development of frontier AI.
Hubinger's warning highlights a growing divide between the rapid progress of AI capabilities and researchers' understanding of how to guarantee that future systems remain safe and controllable.
