OpenAI Safety Chief Departs as Company Merges Research and Safety Teams

OpenAI is losing another leader responsible for the safety of its models. According to a Wired report published on July 11, Johannes Heidecke is stepping down as head of safety systems, and his team will stop operating as a separate division, moving instead under the company's research organization.

Heidecke joined OpenAI in 2021 as an AI safety researcher and took over as head of safety systems in 2024, succeeding Lilian Weng. His work covered model alignment, rule-based reward systems, and readiness evaluations conducted before models were deployed.

Safety Folded Into Research

Rather than functioning as an independent division overseeing researchers' work, the safety team is being folded directly into OpenAI's research organization. That organization is now led by Mia Glaese, who has taken on the newly created role of vice president of research and safety. Saachi Jain has stepped in as interim head of the safety systems team until a permanent replacement for Heidecke is named.

This is not the first departure from OpenAI's safety division in recent months. Andrea Vallone, another safety leader, left the company in late 2025. Before that, in May 2024, Jan Leike, co-lead of the superalignment team, departed OpenAI, and even earlier the company lost co-founder and chief scientist Ilya Sutskever.

Arguments For and Against

OpenAI frames the reorganization as a way to tie safety work more closely to model development, so risk considerations are built in from the start of the process rather than checked at the end as a separate review stage. Critics see the opposite risk: that safety loses its independent voice and becomes subordinate to a team whose primary goal is shipping products faster.

It's important that our safety work be integrated with frontier model development, with an earlier and more direct role in shaping key decisions about models, products and launches - Mark Chen, Chief Research Officer at OpenAI

A Recurring Pattern

Heidecke's departure fits a broader industry pattern in which leaders responsible for model safety regularly leave top AI labs, often shortly after public disputes over the pace of deploying new systems versus the pace of testing them. Market observers note that similar restructurings could become a template for other companies competing for the lead in the model race.

For readers in Poland and the EU, the change carries practical weight, as it coincides with further obligations under the EU's AI Act taking effect, including penalties for developers of the most powerful models set to apply from August 2. How OpenAI organizes its internal safety oversight will shape how the company meets these obligations toward regulators in Europe.

OpenAI has not yet said when it will name a permanent successor to Heidecke or whether further changes to the structure of the teams responsible for assessing model risk are planned. The company maintains that the reorganization will improve collaboration between researchers and the safety team, though the effects of the change will only become clear with future model launches.

Sources: Wired via Bloomberg (bloomberg.com), CryptoBriefing (cryptobriefing.com), TechBuzz AI (techbuzz.ai)