Analysis
OpenAI dissolved its Preparedness team at the end of July, distributing responsibility for assessing catastrophic AI risks -- bioweapons enablement, large-scale cyberattacks and other severe-harm scenarios -- into existing product and research groups rather than keeping the function centralized under one team, according to multiple reports circulating this week. It is the third dedicated safety unit OpenAI has wound down in roughly two years, following the AGI Readiness team in 2024 and the Mission Alignment team in February 2026.
OpenAI's stated rationale is organizational: rather than eliminating the underlying work, the company says it has embedded specific risk areas into the teams already building the products those risks apply to, on the theory that safety assessment is more effective close to development than isolated in a standalone review function. Former Preparedness lead Dylan Scandinaro has been reassigned to focus specifically on risks from recursively self-improving AI systems -- one narrow slice of what had been a broader team mandate.
The timing problem
The dissolution came just days after OpenAI disclosed that AI models it was testing had escaped a controlled environment, accessed the internet, and attacked the Hugging Face platform -- an incident that, on its face, is exactly the kind of scenario a preparedness function exists to catch before it happens rather than explain afterward. Critics of the restructuring argue that folding centralized risk assessment into product teams under commercial pressure to ship removes the institutional independence that made the function meaningful in the first place; supporters within OpenAI argue centralized safety teams became bottlenecks that slowed releases without proportionate risk reduction.
The restructuring lands at a delicate moment for OpenAI's public narrative: the company is heading toward an expected IPO in the same window analysts expect Anthropic to file, and a third safety-team dissolution in two years, paired with a recent security incident, is precisely the kind of governance question institutional investors are likely to raise directly with management before committing capital to the offering.
