OpenAI Safety Head Johannes Heidecke Departs as Company Reshuffles Research and Safety Teams

Johannes Heidecke, OpenAI's head of safety systems, told staff this week that he is leaving the company, WIRED has learned. The departure was confirmed in an internal memo from chief research officer Mark Chen, which detailed a broader reorganization folding safety teams into the research division. Heidecke's exit is not an isolated event. According to CNBC, OpenAI also disbanded another safety team in a separate move, and a head advisor resigned concurrently. The timing points to internal restructuring rather than a single personal decision.

Gary Marcus, writing on his Substack, characterized the situation as "chaos and tension at OpenAI," framing the departure as part of a pattern of leadership churn. The Hill reported that multiple OpenAI executives have departed amid ongoing discussions about a for-profit restructuring, suggesting Heidecke's exit fits a larger organizational transformation. One source, the Latenode community, claimed Heidecke made an alarming statement about being "really scared," though this could not be independently verified by more established outlets covering the story.

The memo from Mark Chen, seen by WIRED, outlines a significant consolidation: OpenAI's safety teams will now report to Mia Glaese, the company's VP of research and head of alignment. Glaese takes on an expanded role as VP of research and safety. Saachi Jain, who previously held a safety-related position, is also mentioned in the reorganization, though her exact new responsibilities remain unclear from the available reporting.

This structural change means safety functions no longer operate as a standalone vertical with their own leadership reporting independently. Instead, they sit under the research organization. Bloomberg notes the reshuffle directly preceded Heidecke's decision to leave, implying a causal link. The Ground publication framed the move as safety teams now reporting to the VP of research, characterizing it as part of a "leadership exodus" reshaping the company. For employees and external observers, the key question is whether integrating safety into research strengthens alignment work by embedding it closer to model development, or dilutes its independence by removing a dedicated safety chief with organizational authority.

Heidecke's exit is the latest in a series of high-profile departures from OpenAI. CNBC reported that the company disbanded another safety team alongside the resignation of a head advisor, though the advisor's name was not specified in available excerpts. The Hill connected these departures to reports of a for-profit restructuring, suggesting that OpenAI's evolving corporate structure is driving leadership changes across multiple functions.

Gary Marcus's Substack analysis described the atmosphere as one of "chaos and tension," arguing that repeated exits from safety-focused personnel indicate systemic issues rather than coincidental timing. The Cryptobriefing noted the departure occurred "amid reorganization," reinforcing the structural rather than personal framing. While no source provided a complete list of recent departures, the cumulative picture from multiple outlets suggests a company in significant organizational flux, with safety leadership particularly affected. The pattern raises questions about OpenAI's ability to maintain institutional knowledge in safety as it pushes toward more advanced models.

OpenAI has publicly committed to prioritizing safety alongside capability development, but the reorganization and Heidecke's exit complicate that narrative. Embedding safety teams under research leadership could accelerate the practical implementation of safety measures by reducing bureaucratic friction between teams. It also removes a dedicated executive whose sole mandate was safety systems, potentially reducing the organizational weight of safety concerns in strategic decisions.

The Latenode community source, while lower in reliability, surfaced a claim that Heidecke expressed fear about the direction of the company, stating "I'm really scared." If accurate, this would signal internal alarm from a departing safety leader. However, no major news organization corroborated this quote, and it should be treated with caution. The Singapore Conference on AI listing for Heidecke confirms his recognized expertise in the field, making his departure a loss of credentialed safety leadership regardless of the circumstances. For the broader AI safety community, the restructuring pattern at OpenAI serves as a case study in how rapidly scaling AI companies balance safety governance against development velocity.

The departure drew attention from major financial and technology outlets. Bloomberg's coverage amplified the WIRED report to its business audience, signaling that leadership instability at the industry's most prominent AI company carries market significance. Techbuzz and Cryptobriefing picked up the story for their respective tech and crypto-focused audiences, indicating broad interest beyond the core AI safety community.

Competitors like Anthropic, which has positioned itself as safety-forward, may see an opportunity to attract talent and credibility as OpenAI navigates organizational turbulence. The Ground's framing of a "leadership exodus" suggests the departures are substantial enough to affect OpenAI's competitive positioning. No source reported direct competitor responses, but the talent market for AI safety expertise is small and intensely competitive. Every departure from OpenAI's safety ranks potentially strengthens rival labs' ability to claim superior safety commitments, a dynamic that matters for enterprise customers and regulatory positioning alike.

Mia Glaese's expanded role as VP of research and safety places her at the center of OpenAI's reconfigured safety apparatus. Her background as head of alignment suggests continuity in technical safety research, but the organizational change means she now manages both capability advancement and safety oversight simultaneously. The effectiveness of this dual mandate will become clear only through the company's actions on future model releases and safety evaluations.

The memo from Mark Chen, as reported by WIRED, indicates the restructuring is already in effect. No timeline was provided for additional changes or hires. Given the concurrent disbanding of another safety team reported by CNBC, OpenAI appears to be streamlining rather than expanding its safety organization. External observers, including regulators and safety advocates, will watch closely for how the company handles safety disclosures and incident reporting under the new structure. The Hill's connection to for-profit restructuring suggests further organizational changes may follow, potentially affecting governance structures that oversee safety decisions.