A warning from inside Anthropic
Evan Hubinger, a former Ripple engineer and current Anthropic researcher, says there is more than a 10% chance artificial intelligence could wipe out humanity in the next decade. He made the estimate while responding to Jacob Coxon, a former Anthropic researcher who resigned Wednesday. Coxon accused leading AI labs of racing toward self-improving superintelligence without adequate safeguards.
On X, Hubinger backed Coxon’s concern. He wrote that the company really does believe AI could kill all humans and put his own estimate above 10% within 10 years. But he said this was not an accusation that Anthropic is ignoring the problem. Anthropic is trying its best, he added, but no plan yet exists to solve alignment for superintelligence, and the lab is not clearly on track.
Hubinger’s job gives the warning unusual weight. At Anthropic he works on alignment stress-testing, checking how safety techniques might fail. This is perhaps why the comments spread quickly: the person flagging the risk is paid to look for cracks in the safety work.
Coxon: labs are gambling with lives
Coxon was more direct. He said neither OpenAI nor Anthropic is acting responsibly. Both are racing straight to self-improving superintelligence and gambling with our lives. He warned that frontier systems might soon exert meaningful influence outside controlled environments. In his view, they will become superhuman systems that can hack anything, change any field overnight, and gain real power and resources. Coxon also said major labs are not properly disclosing AI-related risks to the public. Still, he remains optimistic that top labs could coordinate to slow the race.
The timing is sensitive. The warnings came days after OpenAI and Anthropic released some of their most powerful models. OpenAI introduced GPT-6 Astra on Sept. 3. Nvidia CEO Jensen Huang described it as the arrival of superintelligence. Anthropic released Claude Fable 5.1 and Claude Mythos 5.1, which it calls its most advanced models for coding and scientific research.
From Interledger to AI safety
Hubinger’s path to AI safety did not start in an AI lab. He spent about a year as a software engineering intern at Ripple from June 2014 to August 2015. There he worked on early design of Interledger, a protocol meant to move payments across different ledgers and currencies. He later moved toward theoretical computer science and AI safety, including work at the Machine Intelligence Research Institute.
Ripple CEO Brad Garlinghouse has taken a different public stance on AI. As reported by U.Today, he said Ripple has been adopting AI aggressively and predicted it could boost growth. Garlinghouse called his AI optimism somewhat contrarian. That view now sits against Hubinger’s stark estimate, a reminder that disagreement over AI risk is not limited to one company or one camp.
