John Schulman, a member of OpenAI’s founding team who played a key role in its product development efforts, is leaving for rival Anthropic PBC.

Schulman announced his resignation today in a post on X. His departure reduces the number of OpenAI co-founders still at the company to three: Chief Executive Sam Altman, President Greg Brockman and researcher Wojciech Zaremba. A few hours after Schulman’s announcement, Brockman shared in a separate X post that he is taking an extended leave.

Schulman joined OpenAI in 2015 as a member of its founding team shortly after receiving his computer science doctorate from the University of California at Berkeley. At the university, his research focused on reinforcement learning, a widely used method of training artificial intelligence models. He went on to play a key role in building ChatGPT.

Schulman co-led OpenAI’s post-training team, which is tasked with refining new AI models before the company integrates them into ChatGPT and its other products. The post-training phase of AI projects often focuses on reducing the hardware requirements of a new neural network to reduce inference costs. Researchers sometimes also make other improvements, such as safety upgrades designed to lower the risk of inaccurate AI output. 

Schulman’s other responsibilities at OpenAI included overseeing its AI safety efforts as head of alignment science. In machine learning, alignment is the process of equipping a neural network with safeguards that ensure it won’t generate inaccurate or harmful content.

“I’ve made the difficult decision to leave OpenAI,” Schulman wrote in his X post. “This choice stems from my desire to deepen my focus on AI alignment, and to start a new chapter of my career where I can return to hands-on technical work.”

Anthropic, the competing AI developer that Schulman is joining, has raised more than $7 billion in funding from Amazon.com Inc. and other backers. The company develops a line of large language models known as the Claude series. Anthropic claims that its latest and most advanced model, Claude 3.5 Sonnet, can outperform OpenAI’s flagship GPT-4o LLM across several benchmarks.

Anthropic previously recruited another OpenAI researcher, Jan Leike, who likewise held a leadership role in the latter company’s AI alignment group. At the time, Leike wrote that his team at OpenAI “has been sailing against the wind.” Shortly after his departure, the ChatGPT developer launched an AI safety and security committee that included Schulman.

A few hours after Schulman announced his departure on X, Brockman revealed that he will step back from his role for four months. “I’m taking a sabbatical through end of year,” Brockman wrote. “First time to relax since co-founding OpenAI 9 years ago. The mission is far from complete; we still have a safe AGI to build.”

Previously, The Information reported on Thursday that OpenAI product manager Peter Deng has left the company. Deng joined the ChatGPT developer in 2023. He earlier held roles at Meta Platforms Inc. and other major tech firms.

Image: OpenAI