A prominent safety researcher at Anthropic has sparked intense debate across the tech world after suggesting there is more than a ten percent chance that artificial intelligence could lead to the extinction of the human race within the next decade. Evan Hubinger, who specializes in AI alignment, shared these concerns on X, noting that while current models pose a low immediate threat, the speed of advancement is alarming. He expressed worry that technology could soon reach a tipping point where it improves itself autonomously, creating an existential risk that humanity is not yet equipped to handle.
The warning comes amid growing tension between those developing these tools and those tasked with keeping them safe. Hubinger admitted that despite his company’s efforts, there is currently no clear plan to solve the problem of superintelligence alignment, which involves ensuring high-level AI adheres to human values. This sentiment was echoed by former employee Jacob Coxon, who recently left Anthropic after also spending time at OpenAI. Coxon claimed neither company is acting responsibly as they rush toward creating superhuman systems capable of hacking any network or acquiring independent power and resources.
Not everyone views these dire predictions through a lens of genuine fear. Dame Wendy Hall, a computer scientist advising the United Nations, expressed shock at the public nature of these claims but suggested they might partially be driven by PR and marketing strategies as companies eye lucrative stock market debuts. She questioned why anyone would invest in organizations whose own experts describe their products as potentially species-ending. Meanwhile, reports indicate that Anthropic may have withheld its newest model from the UK’s AI Safety Institute, adding fuel to suspicions regarding transparency in the industry.
This internal friction reflects a broader shift in the global conversation surrounding AI. The debate has moved away from whether these systems pose a danger and toward quantifying exactly how large that danger is. With recent instances of autonomous AI agents conducting cyber attacks and calls for extreme caution from leaders at both OpenAI and Anthropic, many insiders are now pleading for a slowdown in development to ensure humans remain firmly in control of their own future.