"The world is in peril": Anthropic’s head of AI safety resigns, unable to reconcile his work with his values

Mrinank Sharma, who led the Safeguards Research team at OpenAI competitor Anthropic, has left the company. In a farewell letter posted on X – viewed more than ten million times and drawing thousands of comments – he thanked the company for the opportunity to advance AI safety. At the same time, he wrote that growing economic pressure has led to decisions he can no longer reconcile with his values.
In his farewell letter, Sharma not only outlines the reasons for his departure but also delivers a stark warning: “The world is in peril.” He describes the current situation as a “poly-crisis,” in which multiple interconnected crises are unfolding simultaneously. According to him, this is already happening. He makes clear that his concerns are not limited to technical threats such as artificial intelligence or bioweapons, but extend to a broader range of interconnected global crises – including rising geopolitical tensions.
Sharma emphasizes how difficult it is to uphold one’s values in day-to-day work. Even organizations with strong principles face constant pressure to shift priorities or make pragmatic compromises. Anthropic presents itself as a safety-focused, values-driven alternative to other leading AI companies – making his criticism all the more significant.
Sharma plans to study philosophy – community responds with respect
It is notable that Mrinank Sharma plans to leave the tech industry altogether and study philosophy instead. Overall, the community on X has responded with respect and understanding. Many users praise his work in AI safety and wish him well. Some critics argue that he might have had greater influence had he remained at the company. Sharma counters that being outside the organization gives him more freedom to speak openly. This suggests he may continue to follow developments in the tech sector and share his views from a distance.
Beyond his remarks on global crises, Sharma’s farewell letter also contains references that have drawn attention. In the footnotes, he cites a work on so-called “CosmoErotic Humanism,” published under the pseudonym David J. Temple. The name is reportedly linked to Marc Gafni, a spiritual teacher who has faced criticism in the past over allegations of sexual misconduct. While Sharma does not comment on this connection, the reference has raised questions about the intellectual influences behind his decision to step down.
It remains unclear what impact Sharma’s resignation will have on Anthropic’s future. The company behind the chatbot Claude is currently running a major advertising campaign, promoting its AI assistant as an ad-free alternative. In that context, the timing of his departure could prove inconvenient.
Source(s)
MrinankSharma via X (formerly Twitter)
Image source: PIRO4D/Pixabay






