Anthropic insiders admit AI extinction risk exceeds 10% as safety lead confirms no alignment plan

Key points

• Evan Hubinger, an Anthropic alignment lead, stated there is a greater than 10% probability of human extinction via AI within the next decade.
• Jacob Coxon resigned from Anthropic, citing a lack of safety protocols and a reckless race to develop superhuman systems.
• Hubinger admitted Anthropic does not currently have a plan to solve alignment for superintelligence and is not clearly on track to develop one.
• Coxon argued that while he is optimistic about potential coordination among US labs, a global AI race remains inevitable and dangerous.
• The warnings coincide with anticipated IPOs for major AI firms and recent incidents involving rogue AI agents breaching security platforms.

Background

This development follows a series of safety concerns in 2026, including warnings from Israeli firm Irregular about autonomous cyber risks and theoretical studies on 'robotoid humanness.' Previous coverage noted that multiple researchers have left OpenAI over similar safety concerns, and Anthropic was founded by former OpenAI members specifically to address these issues. The current incident highlights a recurring pattern of internal dissent regarding the speed of AI development versus safety measures.

How outlets are covering it

The Verge and CNBC focused on the direct admission by Hubinger regarding the 10% risk threshold and the lack of an alignment plan. The Daily Beast emphasized the broader industry consensus, noting that three researchers at Anthropic issued the warning, and highlighted Coxon’s claim that executives privately express fear despite public reassurances. Yahoo Finance framed the event as a high-profile exit, noting Coxon left AI entirely. While all sources agree on the core facts, The Verge and CNBC highlighted the immediate lack of a safety plan, whereas The Daily Beast focused on the discrepancy between public statements and private fears among executives. CNBC also noted that Anthropic and OpenAI did not immediately respond to requests for comment, contrasting with the public statements made by individual researchers.

Why it matters

The explicit admission by a senior safety researcher that the company lacks a plan to prevent AI misalignment signals a critical gap in governance for the most powerful AI systems. As major labs approach public listings, these internal warnings suggest that financial incentives may be prioritizing speed over safety, potentially increasing the risk of uncontrolled recursive self-improvement. The resignation of a key researcher underscores the growing tension between competitive pressure and ethical safeguards within the AI industry.

What to watch

Regulators and policymakers in the US and EU may face increased pressure to implement stricter oversight or temporary bans on model capability improvements. Investors may scrutinize the safety protocols of AI companies ahead of their anticipated IPOs. Further resignations or public statements from other AI labs could follow as the debate over the feasibility of aligning superintelligent systems intensifies.