A senior artificial intelligence safety researcher has left Anthropic, warning in a public resignation letter that the world faces mounting danger and that companies struggle to let their values guide their actions.
Mrinank Sharma, who led Anthropic’s safeguards research team, announced his departure Monday in a post on X. The letter quickly drew attention and surpassed 1 million views within hours. He wrote that he felt compelled to move on after confronting ethical tensions inside the company.
Sharma said the world faces serious risks, not only from artificial intelligence but from a web of overlapping crises. He suggested that organizations often find it difficult to act in line with their stated principles. However, he did not cite specific incidents or decisions at Anthropic.
He wrote that teams repeatedly face pressure to set aside what matters most. He argued that humanity’s technical power now outpaces its collective wisdom. Consequently, he warned that society may face consequences if moral judgment does not grow alongside technological capability.
Sharma joined Anthropic in August 2023 after earning a doctorate in machine learning from the University of Oxford. He helped launch and lead the company’s safeguards research team last year. The group studied ways to reduce risks linked to advanced AI systems.
Additionally, his team focused on preventing AI-assisted bioterrorism and other malicious uses. Researchers also examined “AI sycophancy,” where chatbots flatter users excessively and reinforce their views. Sharma said the team built systems to detect and block dangerous prompts.
In a report published in May, the safeguards group described efforts to prevent misuse of chatbots. The research examined how individuals might seek guidance for harmful activities.
Read more: Global AI safety review flags deepfakes, job risks, and growing emotional reliance
Read more: Why OpenAI hasn’t yet delivered a traditional return on investment
Sharma did not accuse Anthropic of midconduct
Last week, Sharma released a study examining how chatbots may distort users’ perception of reality. He found that thousands of interactions each day could nudge users toward skewed views. Severe cases remain uncommon, but they appear more often in discussions about relationships and wellness.
Sharma did not accuse Anthropic of specific misconduct. However, his remarks point to broader tensions within the fast-growing AI sector. Consequently, his departure adds to ongoing debates about how companies balance rapid development with ethical responsibility. Meanwhile, governments and researchers continue to debate how to regulate systems that evolve faster than oversight mechanisms.
“We appear to be approaching a threshold where our wisdom must grow in equal measure to our capacity to affect the world, lest we face the consequences,” Sharma wrote in his letter.
He referred to these effects as “disempowerment patterns,” where users rely too heavily on AI advice. The study argued that designers must create systems that protect human autonomy. Additionally, he called for AI tools that encourage independent thinking and healthy decision-making.
Sharma wrote that he may pursue a poetry degree after leaving the company. He said he wants to practice what he described as courageous speech. Meanwhile, he added that he hopes to contribute in ways that align fully with his sense of integrity.
Anthropic operates as a private company and does not trade publicly. It competes with firms such as Alphabet Inc. (NASDAQ: GOOGL), which backs AI research through Google.
Read more: Elon Musk insists the next frontier for AI is above the atmosphere
Read more: Elon Musk claims OpenAI and Microsoft owe him billions from early backing
Part of a longstanding growing trend
Several high-profile departures began during internal disputes at Google over AI safety research and bias. Additionally, former employees said management pushed products forward faster than internal review processes could manage. At Microsoft Corp. (NASDAQ: MSFT) and its partner OpenAI, some staff raised concerns about military contracts and rapid model releases. However, company leaders argued that responsible scaling requires real-world testing alongside safeguards.
Employees at Meta Platforms Inc. (NASDAQ: META) have also questioned the societal impact of large language models and recommendation systems. Meanwhile, whistleblowers and former team members have warned that competitive pressure reduces time for safety checks.
In many cases, departing researchers cited fears about misinformation, deepfakes, and automated decision systems. Consequently, they said companies must slow deployment until oversight frameworks catch up. Some former staff members formed independent research groups focused on alignment and long-term risk. Furthermore, others joined nonprofit labs that prioritize transparency over rapid commercialization.
Executives across the industry have defended their strategies. They argue that strong internal review boards and red-team testing reduce harm. Critics inside these firms often disagreed with that assessment. They said profit incentives can conflict with precaution.
At Amazon (NASDAQ: AMZN), employees publicly criticized the sale of AI tools to law enforcement agencies. Additionally, internal petitions called for clearer rules governing surveillance applications. Industry turnover data suggests ethical friction has become a recurring pattern. However, many employees still choose to remain and influence policy from within.
Recruiters report that experienced AI researchers now ask detailed questions about governance structures before accepting offers. Consequently, ethical positioning has become a competitive factor in hiring.
