Evan Hubinger, who leads the AI safety team at Anthropic, has expressed the view that there is a more than 10% probability that AI will "kill all humans within 10 years." He expressed concern that the development of self-improving AI is progressing faster than expected.
SecurityAnthropicEvan Hubinger
Anthropic Safety Researchers Express Concern that AI Could Cause Human Extinction Within 10 Years with Over 10% Probability
This article is a translation. Read the Japanese original
Additionally, Jacob Coxon, a researcher who left Anthropic, stated on X (formerly Twitter) his concerns that safety measures are insufficient, alleging that the company and OpenAI are engaged in a development race over uncontrollable superintelligence. Coxon argued that the development race is being prioritized despite the fact that people involved in AI development seriously believe that "AI could kill all humans within 10 years."
According to Hubinger, plans to align advanced AI with human values and ensure safety have not been established at this time. Within the industry, long-term concerns have been raised regarding the risks of "recursive self-improvement," where AI repeatedly improves itself beyond human control.
Sources
- Worried Anthropic researchers warn that AI ‘could kill all humans’ （The Verge AI, 2026-09-09）