AI safety researcher quits Anthropic, warns of global risks

An AI safety researcher, Mrinank Sharma, resigned from US-based Anthropic, citing concerns about global risks and the state of the world.

Sharma, who led a team focused on AI safeguards, said he plans to pursue poetry and move back to the UK to “become invisible.”

He expressed worries about interconnected crises beyond AI, including bioweapons, and criticized the difficulty of aligning corporate actions with core values.

Anthropic, founded in 2021 by former OpenAI employees, has faced scrutiny over its practices, including a US$1.5 billion settlement related to training data.

The company has released reports on AI safety and engaged in public disputes with competitors like OpenAI over ads and ethical concerns.

🔗 Source: BBC

🧠 Food for thought

Implications, context, and why it matters.

This resignation matters because of who the researcher was and what Anthropic claims to be

Mrinank Sharma led Anthropic’s AI safeguards work, including the Safeguards Research Team that worked on defenses against catastrophic misuse such as AI-assisted bioterrorism 1.
His warning about the gap between corporate decisions and stated values carries weight because Anthropic is a public benefit corporation (PBC), which is intended to pursue a public-benefit purpose alongside shareholder value 2.
Anthropic was founded in 2021 by former OpenAI staff. The company also set up a Long-Term Benefit Trust in its governance to balance investor interests with its public-benefit mission 3.
The exit suggests that even this governance model can buckle under commercial pressure to keep up with rivals such as OpenAI and Google 1.

The departure flags a shift in defining AI harm and casts doubt on corporate self-governance

Sharma’s last project moved beyond bioweapons to a quieter risk. He argued that AI assistants could make people “less human” or warp their humanity 4.
That framing pushes safety talk past one-off disasters to slow social damage, where regulators have fewer tools.
The departure fits a wider run of safety-linked exits across large AI labs. Some observers read those exits as friction between safety work and the rush to ship stronger models, though his letter focused on Anthropic plus broader global crises 1.
For regulators and customers, the resignation serves as a warning that safety branding should not be accepted on trust, even at firms like Anthropic that lean on safety and public-benefit goals 5.