GPT-5.6刚发布，OpenAI安全主管就跑路了？？

On July 11, 2026, OpenAI safety lead Johannes Heidecke announced his resignation. WIRED confirmed the news, stating that his departure comes as the company seeks to further integrate its research and safety teams. Heidecke's exit is not an isolated incident: just prior, OpenAI released GPT-5.6, and the safety lead's resignation closely following the product launch is a telling timing.

A report from Quantum Bit more directly describes this as "the sixth safety lead to leave in two years," noting that Heidecke's resignation is the latest footnote in OpenAI's long-standing turmoil in safety governance. The internal tension between safety and speed has amplified with each high-level departure.

Heidecke is the sixth safety lead to leave in two years. Previously, heavyweights like Miles Brundage and Jan Leike departed due to disagreements with the company over safety priorities. These resignations are often accompanied by public statements or internal letters that reveal OpenAI's oscillation between commercial profitability and safety responsibility.

The frequent turnover in the safety team is not accidental. Since OpenAI's transition from a non-profit to a commercial entity, product release speed has accelerated, and safety review processes have been compressed. Multiple former safety executives have pointed out that the company's safety culture is being eroded by a culture of "moving fast."

Notably, not all departures are "forced." Some executives chose to leave after feeling powerless to advance the safety agenda. This suggests that OpenAI's internal safety governance structure may suffer from systemic issues rather than mere personal conflicts.

OpenAI's mission statement has always emphasized "ensuring AGI benefits all of humanity," but in practice, product iteration speed often takes precedence over safety justification. The rapid release of GPT-5.6 accompanied by the safety lead's departure highlights this contradiction.

The integration of safety and research teams was intended to bring safety considerations earlier into the R&D process, but its actual effect has been the loss of safety leads. Heidecke's resignation may indicate that the integration failed to address the root problem—namely, the safety team's insufficient voice in key decisions.

Some analysts believe that OpenAI's safety exit wave is not an isolated case but a structural challenge facing the entire AI industry: when commercial competition intensifies, safety is often the first to be sacrificed. However, as the industry leader, OpenAI's safety turmoil has a stronger demonstration effect.

The frequent turnover of OpenAI's safety leads may undermine public and policymaker confidence in AI safety. Regulators are closely monitoring AI companies' governance practices, and instability in the safety team could become a catalyst for stricter regulation.

Additionally, the outflow of safety talent may flow to competitors or independent research institutions, potentially exacerbating overall industry safety risks. If the top AI safety experts cannot have impact within OpenAI, they may establish new safety standards elsewhere, but they could also see reduced influence due to resource fragmentation.

In the short term, Heidecke's departure will not immediately affect GPT-5.6's operations, but long-term safety culture rebuilding requires stable leadership and institutional safeguards. Whether OpenAI can attract and retain safety talent in the future will be a litmus test of its safety commitment.

Six safety leads in two years—this number alone is a harsh indictment of OpenAI's safety governance. Heidecke's departure is not an endpoint but a warning: when a company grows rapidly, safety cannot be just a slogan.

For OpenAI, the first step to rebuilding a safety culture is acknowledging the problem and giving the safety team real decision-making power. Otherwise, the departure of the GPT-5.6 safety lead might just be the beginning of the next wave of exits.

For the entire AI industry, this event should prompt all companies to reflect: while pursuing performance breakthroughs, how to ensure that safety mechanisms are not just decorative. After all, AI safety issues will not disappear with personnel changes; they will only await the next crisis to erupt.