OpenAI co-founder Schulman jumps ship to Anthropic

The brain behind training ChatGPT to actually follow instructions is heading to the competition.

OpenAI co-founder John Schulman just announced he is leaving the company to join rival startup Anthropic. At the same time, OpenAI confirmed that president and co-founder Greg Brockman is taking an extended leave through the end of the year to recharge, while top product executive Peter Deng has also quietly exited.

With Schulman out, only three of OpenAI's original 11 founders remain: CEO Sam Altman, Brockman (now on leave), and language lead Wojciech Zaremba.

The post-training mastermind switches teams

Schulman joined OpenAI straight out of UC Berkeley back in 2016 and led the reinforcement learning and post-training teams that turned raw models into ChatGPT. Basically, he is the reason generative AI talks like a cooperative assistant instead of spewing gibberish.

After OpenAI disbanded its Superalignment group less than three months ago, Schulman took over alignment science and joined the company's new safety committee. Now, he is reuniting with former colleague Jan Leike at Anthropic.

In his departure note on X, Schulman tried to keep things cordial, insisting he was not leaving due to a lack of support for alignment research from leadership. Instead, he framed the jump as a personal choice to return to hands-on technical work alongside researchers deeply focused on safety.

Altman thanked Schulman publicly, calling him a brilliant researcher and great friend. Leike, meanwhile, replied directly to Schulman's post: "Very excited to be working together again!"

Executive shuffle continues

Schulman's departure is not happening in a vacuum.

Greg Brockman, OpenAI's operational backbone, announced a months-long sabbatical after nine nonstop years. Meanwhile, VP of Product Peter Deng, a heavy hitter hired last year after leadership stints at Meta, Uber, and Airtable, has also departed.

Since the boardroom coup attempt against Altman last November, OpenAI has experienced steady executive bleeding. Co-founder Ilya Sutskever left earlier this year to launch Safe Superintelligence (SSI), and now Schulman is gone too.

Why it matters

Anthropic has essentially turned into the premier destination for OpenAI's disillusioned alignment pioneers. Securing Schulman is a massive technical win for the Amazon-backed lab. Post-training and RL are arguably the hardest, most secretive parts of modern frontier models, and Anthropic just acquired the architect who built ChatGPT's pipeline.

For OpenAI, the optics look increasingly concerning. Altman continues to insist that safety remains paramount, touting partnerships with the US AI Safety Institute and pledging 20% of compute to safety initiatives. But actions speak louder than press releases. When foundational researchers repeatedly leave to do "hands-on alignment work" elsewhere, it signals that OpenAI's intense commercial race may no longer leave room for pure, cautious science.