Karla Barker 
OpenAI has quietly broken up one of its main internal safety groups. The preparedness team, which existed to catch the most dangerous risks in AI models, was disbanded at the end of July 2026, according to the Financial Times. The unit had a blunt job: ask whether a model could do real damage before it ships. That meant testing for possible misuse in areas like bioweapons development and large-scale cyberattacks, then designing ways to reduce those risks.
What happened to the preparedness team
That centralized function is gone. Responsibility for bio and cyber risk has been split up and folded into existing teams. There is no longer a single group whose sole purpose is to watch for catastrophic misuse across the board. Whether this distributed setup catches the same problems just as clearly is hard to verify from the outside. OpenAI has not offered a detailed public explanation. The Financial Times reported the move came alongside broader restructuring as the company heads toward a large expected IPO.
Not the first safety team to go
This is not isolated. OpenAI has steadily moved away from the long-horizon safety structure it once built its reputation on. It previously shut down its AGI readiness team and its superalignment team, both focused on preparing for AI systems that become dramatically more capable than today’s models. Taken together, those decisions point to a company narrowing its safety apparatus while models get stronger. Any single dissolution might look routine. Stacked one after another, it looks more like a retreat from the long-term risk planning OpenAI used to talk about openly.
The restructuring also came alongside exits. Ethics lead Chloé Bakalar, Chief Futurist Josh Achiam, and head of safety Johannes Heidecke have all left recently. None of them, as far as public records show, were replaced by someone with the same title. Jan Leike, who left OpenAI in 2024, told the Financial Times the company was ignoring safety in favor of shiny products. That accusation lands harder after so many safety roles have emptied out.
Why the timing matters
The changes come as OpenAI moves toward what is expected to be a massive initial public offering. Companies preparing for that kind of scrutiny tend to streamline and cut anything that doesn’t fit a clean growth story for investors. A dedicated risk team focused on worst-case scenarios may not fit that narrative, even if it fits the mission OpenAI once claimed. This matters beyond OpenAI. The company has long positioned itself as a leader in responsible AI development. If a company at this scale folds dedicated risk teams into product groups ahead of a public listing, regulators and competitors are likely to take note.
Who is steering risk work now
Dylan Scandinaro, who led the preparedness team before it was dissolved, is not leaving. He was hired from Anthropic in February. He is now focused on a narrower but perhaps more extreme question: the implications of recursive self-improving AI, or systems that can improve their own capabilities without much human involvement. That suggests OpenAI still wants people watching its most extreme long-term risks, just not organized the same way. Whether splitting bio and cyber risk into product teams adds up to equal coverage remains the open question. For now, OpenAI has not said publicly how the redistributed duties will be measured the way a standalone team once was.