WASHINGTON: Artificial intelligence is moving forward so quickly that a senior safety researcher at Anthropic says there is more than a 10% chance it could wipe out humanity in the next ten years.
Evan Hubinger, an AI alignment researcher at the Amazon-backed lab, gave this warning soon after another researcher, Jacob Coxon, announced he was leaving the company. Coxon, who worked at OpenAI before joining Anthropic, said top tech labs care more about competing in the market than about safety. He said developers are “racing straight to self-improving superintelligence and gambling with our lives.”
Hubinger said that while current models are not an immediate threat, the fast move toward recursive self-improvement, where models can upgrade themselves, could be very dangerous in the long run.
“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,” Hubinger wrote on X in a post seen by millions. Alignment means making sure artificial intelligence systems reliably follow human values and safety rules.
This disagreement inside the company comes as worries about outside safety checks grow. The Financial Times reported that Anthropic did not share its newest top model with the UK’s AI Safety Institute (AISI), a leading independent safety tester. The UK Cabinet Office said it is still working closely with industry partners, but researchers pointed out that competition with China and other countries is making developers more secretive.
This situation shows a widening split in the artificial intelligence field. Executives are investing large sums and planning public listings, but researchers warn that autonomous AI agents already pose serious risks, including the ability to launch cyberattacks. More scientists in the industry are now asking governments to create international rules to slow down development before these models get out of human control.
