Anthropic's head of AI safety Mrinank Sharma resigns, says 'world is in peril' in resignation letter

Mrinank Sharma, who led the safeguards research team at Anthropic, has stepped down, issuing a thought-provoking resignation letter on X (formerly Twitter) on February 9. In his letter, Sharma cautioned that the world stands at a critical juncture, grappling with a multitude of interconnected crises. He declared, "Today is my last day at Anthropic. I resigned. Here is the letter I shared with my colleagues, explaining my decision." Sharma's resignation comes at a crucial time for the company, known for its safety-oriented approach, as it shifts towards a more commercial model with aspirations of reaching a $350 billion valuation, backed by Amazon and Google. His letter, rich with references to poets like Rainer Maria Rilke and William Stafford, emphasized a worrying trend: humanity's technological advancements are outpacing its ethical considerations. He warned, "We seem to be nearing a threshold where our wisdom must expand in tandem with our ability to impact the world, or we risk dire consequences." The letter also highlighted Sharma's concern about a range of crises affecting the globe, extending beyond the typical anxieties associated with artificial intelligence. His resignation has ignited discussions surrounding Anthropic's internal culture, especially as the company was founded by former OpenAI executives who left due to similar concerns about commercialization. Sharma reflected on the challenge of allowing core values to guide decisions in a rapidly evolving environment, stating, "I've repeatedly seen how difficult it is to genuinely let our values dictate our actions." Notably, one of Sharma's final projects examined the potential for AI assistants to alter human behavior, raising alarms about technology making users 'less human.' This concern arises as the company embraces the development of 'agentic' AI aimed at enhancing workplace productivity. His departure follows closely on the heels of the release of Claude Opus 4.6, a new model intended for advanced coding and office tasks, raising questions about whether the push to deliver quickly for investors may have jeopardized the safety measures his team was responsible for upholding. Sharma's exit is part of a larger trend, as he is not the only prominent figure to leave Anthropic recently; AI scientist Behnam Neyshabur and R&D expert Harsh Mehta also departed last week. As of now, Anthropic has not publicly addressed Sharma's resignation or the pressing issues he raised in his letter.