AI researchers at Anthropic and OpenAI warn of extreme risks

In short: A growing number of AI researchers are publicly warning that advanced AI could cause catastrophic harm if it is built too quickly without better safeguards.

What's going on

Several current and former researchers linked to Anthropic and OpenAI have made unusually direct warnings about the dangers of advanced AI. Some have said the risks could include human extinction if development keeps moving fast without stronger safety work.

Recent attention focused on Anthropic, where former researcher Jacob Coxon said he left because he believed leading AI labs were “gambling with our lives.” Anthropic alignment lead Evan Hubinger said he believes there is a greater than 10 percent chance AI could kill all humans within the next decade, according to reporting. Anthropic researcher Samuel Marks also said developers believe AI could cause “human extinction” or similarly severe outcomes in the near future.

This is part of a broader pattern, not one isolated comment. Reports earlier in 2026 described multiple resignations across major AI companies, tied to fears about fast development and risks like biological threats, manipulative chatbots, and cyber misuse. In August 2026, researchers from OpenAI, Anthropic, Google DeepMind, Meta, and others signed an open letter urging the US government to deliberately pace “frontier AI” development, meaning the most advanced systems.

Behind many of these warnings is a concern that future AI systems may become harder to control or predict. Researchers worry models could get better at planning, deception, hacking, or even improving themselves (like a tool that keeps upgrading its own parts).

What to watch

Watch for whether governments respond with new rules, and whether leading labs slow down or share more details about their safety testing. Also watch the debate between people focused on extreme long-term risks, and those focused on near-term harms like scams, deepfakes, and cyberattacks.

Source: NYTimes