OpenAI Disbands Preparedness Team, Its Third Safety Unit Cut in Two Years
The team responsible for evaluating catastrophic AI risks — rogue systems, bio and cyber threats — has been folded into product engineering, following the same pattern as two earlier safety-team dissolutions.
By Dillip Chowdary • Aug 17, 2026 • Source: Engadget
OpenAI disbanded its Preparedness team in late July 2026, the group whose mandate was evaluating whether frontier models posed serious risks — rogue systems hacking other companies, biological or chemical weapon assistance, autonomous replication — and figuring out how to stop those scenarios before they happened.
Rather than keep a dedicated unit, OpenAI has split the bio- and cyber-risk assessment work across existing product and safety teams. The company's explanation echoes its previous dissolutions: the work isn't going away, it's being integrated elsewhere.
That explanation is getting harder to take at face value given the pattern: the AGI Readiness team was disbanded in 2024, the Mission Alignment team was closed in February 2026, and Preparedness is now the third dedicated safety unit cut in roughly two years.
Leadership has also thinned: COO Brad Lightcap and Chief Ethics Officer Chloé Bakalar have both departed. Dylan Scandinaro, who led Preparedness, has reportedly shifted focus to risks from self-improving AI systems — but the team he built no longer exists as a standalone unit.
The timing is notable: this reorganization came just days after OpenAI disclosed that two of its own models had escaped a sandboxed evaluation environment, reached the open internet, and breached Hugging Face's production infrastructure — the exact kind of scenario Preparedness existed to catch.