Anthropic AI Safety Researcher Resigns, Warns ‘World Is in Peril’

TL;DR: Mrinank Sharma, who led Anthropic’s safeguards research team, resigned publicly with a letter warning that the “world is in peril.” His post on X was viewed over a million times, adding to a pattern of high-profile departures from leading AI companies.

Sharma, who holds a PhD in machine learning from the University of Oxford, joined Anthropic in August 2023. His team worked on defences against AI-assisted bioterrorism and conducted research into AI sycophancy — the tendency of chatbots to tell users what they want to hear rather than what is accurate.

What He Said

In his resignation letter, posted publicly on X, Sharma wrote that the world faces a “whole series of interconnected crises.” He said he had “repeatedly seen how hard it is to truly let our values govern our actions” at Anthropic, describing “pressures to set aside what matters most.” He did not provide specifics about which values were compromised or what form those pressures took.

His team published a study just last week finding that “thousands” of daily AI chatbot interactions may produce distorted reality perceptions in users — a finding that sits uncomfortably alongside the commercial incentive to make chatbots as engaging as possible.

A Broader Pattern

Sharma’s departure follows a series of notable exits from AI safety roles across the industry. Tom Cunningham left OpenAI in September. Jan Leike departed OpenAI for Anthropic before that. Gretchen Krueger also left OpenAI. The cumulative effect of these departures has raised questions about whether safety teams at leading AI companies have the influence their employers publicly claim they do.

Looking Forward

Sharma said he plans to pursue a poetry degree and “devote myself to the practice of courageous speech.” His departure is unlikely to alter Anthropic’s trajectory on its own, but the public nature of his resignation — and the audience it reached — adds to growing scrutiny of how AI companies balance commercial pressures against safety commitments. When the people hired specifically to keep these systems safe start walking away and saying so publicly, the gap between safety rhetoric and internal practice becomes harder to dismiss.