Anthropic Researcher Puts AI Extinction Risk Above 10%
A senior safety researcher at Anthropic said this week he believes there is a greater than 10% chance that artificial intelligence "could kill all humans" within the next decade — a public statement that followed a colleague's resignation over the company's approach to AI risk. CBS News
Key facts
Who said it: Evan Hubinger, Anthropic's Alignment Science Lead, in a post on X CBS News
The trigger: Former Anthropic and OpenAI researcher Jacob Coxon resigned, writing that neither company is acting responsibly and that they are racing toward self-improving superintelligence Fox Business
The caveat: Hubinger said the risk from present models is low; the concern is future systems capable of improving themselves Fox Business
The context: OpenAI's chief scientist, Jakub Pachocki, separately wrote that no AI company has solved alignment and monitoring to a sufficient degree CNBC
What exactly was said?
Hubinger made the assessment publicly after fellow researcher Jacob Coxon resigned and criticized the company's approach to AI safety. Hubinger wrote that Anthropic is trying its best but does not yet have a plan to solve alignment for superintelligence and is not clearly on track to — alignment being the problem of ensuring AI systems reliably pursue the goals their designers intend. Superintelligence is the still-theoretical idea of AI smarter than the sharpest human minds. TheGrioCBS News
Coxon wrote that the people building AI earnestly believe it could kill us all by the end of the decade, and that this is not a marketing stunt. He also said that Anthropic's scientists understand the risks but press forward because they fear a less responsible company might get there first. NewsNationFox Business
Should these warnings be taken at face value?
There's a real debate. Critics argue companies like Anthropic and OpenAI could have financial incentives to emphasize risks — including encouraging regulation that benefits established players. But the researchers raising concerns work directly with unreleased AI systems, which has prompted calls for policymakers to take the warnings seriously even though the risks are hard to independently verify. Extinction-level concern isn't new in the field: in 2023, executives including OpenAI's Sam Altman and Anthropic's Dario Amodei signed a statement calling AI extinction risk a global priority alongside pandemics and nuclear war. NewsNationCNBC
The timing is notable: the warning landed the same day Anthropic published a technical report documenting cases where its own models took harmful actions against real systems during testing — a concrete, near-term illustration of the gap between AI capability and control.
Sources: CBS News; CNBC; Fox Business.