Anthropic Researchers Warn AI Could Kill All Humans by 2030
A departing Anthropic safety researcher warns that AI labs are gambling with human lives by recklessly racing to build superhuman artificial intelligence.
Stock photo for illustration only, not from the actual event
- Jacob Coxon resigned from Anthropic over safety concerns regarding self-improving AI systems.
- Anthropic safety lead Evan Hubinger estimates a greater than 10 percent chance AI could kill all humans within the decade.
- Researchers warn that companies are locked in a race for superintelligence without a clear safety plan.
The artificial intelligence community has been shaken after a senior safety researcher at Anthropic stated there is a greater than 10 percent chance that advanced AI could eradicate humanity by the end of the decade. The warning came hot on the heels of a colleague resigning over fears that AI laboratories are carelessly rushing to build uncontrollable superhuman systems.
In a post announcing his departure on X, Jacob Coxon, an AI training researcher at Anthropic, stated he quit due to the company's lax approach to safety. Coxon, who previously worked on training systems for OpenAI, accused both AI giants of racing directly toward self-improving superintelligence and gambling with human lives, despite builders earnestly believing it could kill everyone by 2030.
Industry insiders have long warned about the dangers of self-improving AI loops running out of human control, a concept known as recursive self-improvement. While this milestone has not yet been fully reached, companies are aggressively pursuing it, and much of the underlying AI code is already written with the assistance of artificial intelligence.
Stock photo for illustration only, not from the actual event
In direct response, Evan Hubinger, who leads an AI safety team at Anthropic, admitted his own concerns over self-improving AI, stating it is happening faster than initially anticipated. He echoed Coxon's assessment, estimating the probability of human extinction from AI to be higher than one in 10 within the next ten years.
"We really do earnestly believe AI could kill all humans"
Evan Hubinger
Despite these existential warnings, Hubinger noted that Anthropic does not yet have a concrete plan to ensure advanced AI remains safe and aligned with human values. Coxon asserted that companies are trapped in a competitive race to deploy advanced systems first, pushing forward regardless of the catastrophic risks involved.
This high-profile departure underscores mounting internal tensions within the tech sector as leading AI firms accelerate their development timelines ahead of anticipated IPOs. It also highlights ongoing challenges with rogue agent incidents and the difficulties in monitoring frontier models, raising urgent policy questions about how society will govern rapidly evolving artificial intelligence.
Coxon's exit marks a significant milestone for Anthropic, a company originally founded by former OpenAI employees who also left due to safety disagreements, illustrating a persistent cultural rift between commercial pressures and existential risk management in the tech industry.
Source: The Verge
Found something wrong in this article? Report an issue with this article
Comments
Leave a Comment