Key AI safety researcher leaves Anthropic
Mrinank Sharma, who led safeguards research at Anthropic, resigned from the AI company yesterday. He made his departure public by sharing the letter he sent to colleagues on social media platform X.
In that letter, Sharma expressed what he called “mounting unease” about the space between what AI organizations say they believe and what they actually do. He pointed to a growing disconnect between ethical commitments and how things work in practice. Not just at companies, but in society more broadly.
“It is clear to me that the time has come to move on,” Sharma wrote in the post that included his resignation letter.
Two years of safety work
Sharma spent two years at Anthropic, the developer of the Claude AI assistant. During that time, his work focused on several critical safety areas. He helped build defenses against potential AI-enabled biological threats. He worked on internal accountability tools. And he contributed to early frameworks for documenting AI safety measures.
His research also looked at how chatbots might reinforce user biases over time. And how they could gradually reshape human judgment through repeated interactions. This is the kind of subtle, long-term risk that doesn’t always get attention amid more immediate concerns.
A shift away from corporate AI
In his letter, Sharma praised his former colleagues. He acknowledged their technical skill and what he called their “moral seriousness.” But his decision to leave signals a broader shift away from corporate AI work entirely.
He announced plans to pursue writing and personal coaching. He’s also considering graduate study in poetry. That’s quite a change from AI safety research, but perhaps it reflects his desire to step back from the field that’s causing him concern.
This resignation comes at a particular moment. There’s been heightened attention recently on how leading AI developers handle internal dissent. People are watching how these companies disclose risks. And there’s ongoing debate about balancing rapid capability gains against adequate safety research.
The broader conversation
Sharma’s departure adds another voice to that conversation. When someone who worked directly on safety measures decides to leave and speak publicly about their reasons, it gets noticed. It makes you wonder what they saw that prompted this move.
I think it’s worth noting that he didn’t just quietly resign. He shared his letter publicly. That suggests he wants this to be part of a larger discussion about AI ethics and safety practices.
The timing feels significant too. As AI capabilities advance quickly, the questions about safety and ethics become more urgent. When researchers working on these issues express doubts about the systems they’re helping to build, perhaps we should pay attention.
It’s not clear what immediate impact this will have on Anthropic’s work. The company has other researchers continuing safety efforts. But losing someone who led safeguards research is notable. Especially when they leave expressing concerns about ethical gaps.
What happens next will be interesting to watch. Both for Sharma’s new direction and for how Anthropic and other AI companies address these kinds of concerns from within their own teams.
