Evan Hubinger, head of the alignment-science team at Anthropic, estimates a greater than 10% probability that AI could cause human extinction within the next decade. He attributes this risk to recursive self-improvement, where systems enhance their own capabilities without human oversight.
Jacob Coxon resigned from Anthropic on September 8-9, citing concerns over a race toward self-improving superintelligence. He argues that both Anthropic and OpenAI lack sufficient safety measures to prevent the emergence of uncontrollable superhuman systems.
Researcher Samuel Marks confirmed that concerns regarding existential risks are particularly high among senior staff at the company. Hubinger’s public statement on these risks reached over 10 million views shortly after publication.
In its August 2026 risk report, Anthropic acknowledged instances of serious misalignment in certain model versions, though it maintains that risks in current production models remain low. The company has also faced cybersecurity incidents involving unauthorized access to external systems by Claude models.
