OpenAI disbanded its Preparedness team at the end of July, eliminating the 3rd dedicated safety group the company has dissolved in roughly two years.
Key Points:
- OpenAI shut its Preparedness team in late July and split biology and cybersecurity risk work across existing teams.
- The change follows a wave of senior safety and ethics departures as the company moves toward a public listing.
- OpenAI says its Preparedness Framework and Safety Advisory Group remain in place, with the same risk categories intact.
OpenAI Dissolves Preparedness Team Before IPO
The Financial Times first reported the change, citing several people familiar with the matter. The unit judged whether frontier models could cause severe or catastrophic harm, then studied ways to blunt those risks.
After the restructuring, the company carved up that mandate by domain and handed biology and cybersecurity work to senior staff scattered across existing teams.
What disappeared was the reporting line rather than the written rulebook that governs how the company escalates dangerous model capabilities. The Preparedness Framework still covers biochemical threats, cyber capability and self-improving models, with oversight from an internal Safety Advisory Group.
Also Read: Bitcoin Faces Greater Selloff Risk After 90 Days Of Weak U.S. Demand
Safety Departures Deepen Scrutiny Of OpenAI
The reorganization arrived during an unusually heavy run of senior exits. Head of safety systems Johannes Heidecke, ethics lead Chloé Bakalar and chief futurist Joshua Achiam all left within weeks of one another. Heidecke's group was folded into a research organization run by Mia Glaese, a vice president overseeing research and safety.
Co-founder Greg Brockman has argued that embedding safety inside model development and product teams strengthens the work, rather than leaving it isolated in one separate unit. Executives have also said they want risk checks applied earlier in training, closer to the engineers who actually build the systems.
Skeptics read the same facts differently. They note that nobody inside the company now owns risk alone, independent of the release calendar and its commercial pressure. Jan Leike, who quit the company in 2024 for rival Anthropic, has said safety there slipped behind "shiny products."
Scandinaro Arrived From Anthropic In February
The move unwinds a bet OpenAI made only six months ago. Its consequences now land inside a business preparing one of the largest technology listings in years.
OpenAI recruited Dylan Scandinaro from Anthropic in February to lead preparedness, and Sam Altman called him the best candidate for the role. Scandinaro warned then of "risks of extreme and even irrecoverable harm." He now focuses on recursively self-improving systems, though the team he was hired to build no longer exists around him.
Read Next: CZ Moves $965K To Giggle Academy As Binance Founder Shifts Focus To Education
