- Published on
OpenAI’s Head of Safety Is Leaving the Company
- Authors
- Name
- DeepHorus
OpenAI’s head of safety systems, Johannes Heidecke, has announced his departure from the company, according to an internal memo obtained by WIRED. Heidecke’s exit marks another significant leadership change at the AI powerhouse, which has been grappling with the challenge of balancing rapid model development with robust safety measures. His departure follows a recent reorganization that aimed to more closely integrate OpenAI’s safety and research teams, signaling a shift in how the company approaches risk management.
The reorganization, detailed in a memo from chief research officer Mark Chen, places OpenAI’s safety teams under the oversight of Mia Glaese, who will now serve as vice president of research and safety. Glaese, previously head of alignment, will take on an expanded role. Saachi Jain, who previously led safety teams, will serve as interim head of safety systems, reporting directly to Glaese. The changes come as OpenAI faces mounting pressure to ensure its increasingly capable models are deployed responsibly.
Chen’s memo highlighted the accelerating pace of AI development: “The demands on safety continue to increase—we are training models at a much faster cadence, and release cycles have come down greatly in turn. As a result, we have bigger coordination challenges around safety today than ever before.” This statement underscores the tension between innovation and oversight, a central issue for the company as it pushes the boundaries of AI.
Heidecke joined OpenAI in 2021 as an AI safety analyst and became head of safety systems in 2024, succeeding Lilian Weng, who left to co-found Thinking Machines Lab. During his tenure, Heidecke oversaw the safety review processes for groundbreaking models like GPT-4 and GPT-4o. His departure, while abrupt, is framed by the company as a natural consequence of organizational evolution aimed at embedding safety earlier in the development lifecycle.
The impact of this leadership shuffle is significant for OpenAI’s internal culture and external reputation. Safety researchers and external watchdogs may view the constant turnover with skepticism, especially as the company continues to release models with known risks. Just this week, OpenAI launched GPT-5.6, its most advanced model for agentic coding, yet acknowledged it exhibited “concerning forms of misaligned behavior.” This admission reinforces the need for robust safety systems—and raises questions about whether the new structure can deliver.
Industry reactions have been mixed. Some analysts applaud the integration of safety and research, arguing that siloed teams often struggle to influence product decisions. Others worry that safety leaders are being marginalized. Chen’s statement expressed gratitude for Heidecke’s contributions and emphasized the importance of integrating safety work with frontier-model development, giving it an “earlier and more direct role” in key decisions.
Looking ahead, Mia Glaese will be under the spotlight as she leads safety across research and product launches. Her expanded role suggests OpenAI is betting on tight coordination between research and safety to avoid public failures. However, the company must also contend with the departure of other key figures: earlier this week, chief futurist Joshua Achiam left after nine years, and CEO of AGI deployment Fidji Simo stepped down following medical leave. Greg Brockman will continue leading product teams and take on go-to-market strategy.
These departures are part of a broader trend at OpenAI, where high-profile exits—including co-founders Ilya Sutskever and John Schulman—have raised concerns about the company’s direction. As OpenAI transitions from a research lab to a product-driven organization, it faces the delicate task of maintaining its safety culture while meeting commercial demands. The new organizational structure, with Glaese at the helm, will be tested as the company races toward artificial general intelligence. Whether this consolidation of authority can stabilize safety leadership remains an open question, but it reflects a recognition that safety cannot be an afterthought in the age of rapidly advancing AI.
