Anthropic snags former OpenAI safety chief Jan Leike

Just weeks after resigning from OpenAI and publicly calling out the company for deprioritizing safety, Jan Leike has found a familiar new home.

The former OpenAI safety lead announced he is joining rival Anthropic to head up a newly formed superalignment team.

Rebuilding OpenAI's dissolved mission

Leike revealed on X that his team at Anthropic will tackle key frontiers in AI safety. Specifically, they will focus on scalable oversight, weak-to-strong generalization, and automated alignment research.

According to sources familiar with the matter, Leike will report directly to Anthropic chief science officer Jared Kaplan. Current Anthropic researchers already working on scalable oversight will transition to report to Leike as the new group ramps up.

If that agenda sounds familiar, that is because it directly mirrors the mission of OpenAI's recently disbanded Superalignment team. Leike previously co-led that initiative, which aimed to solve the control problem of superintelligent AI within four years before running into friction with OpenAI leadership.

A familiar migration pattern

Anthropic has long positioned itself as the safer, more principled alternative to OpenAI. In fact, the company was born out of a similar fracture.

Anthropic CEO Dario Amodei was formerly OpenAI’s VP of research. He famously split from the startup alongside former policy lead Jack Clark and others due to disagreements over OpenAI's accelerating commercial drive.

Now, history seems to be repeating itself as Anthropic welcomes another high-profile OpenAI defector looking for safe harbor.

Why it matters

This is a massive cultural and technical victory for Anthropic. As OpenAI faces growing criticism for sidelining safety in favor of speed and commercial products, Anthropic is actively absorbing the talent OpenAI pushed out.

We think this move shows where the center of gravity for frontier safety research is shifting. If you are a safety researcher who felt disillusioned by OpenAI dissolving its Superalignment unit, Anthropic just made itself the obvious landing spot.

Anthropic played this smartly by placing Leike right under its chief science officer and handing him existing research staff immediately. But the real test is yet to come. High-end alignment research eats tremendous compute, and Anthropic faces the exact same commercial pressures from backers like Amazon. Whether they actually fund Leike's vision better than OpenAI did remains to be seen.