OpenAI disbanded its preparedness team at the end of July 2026, folding the group's work into existing product-area teams as the company moves toward an expected IPO, the Financial Times reported on August 16. The preparedness team's mandate was to assess whether frontier models posed catastrophic risks — including autonomous behaviors like a model attempting to hack another company — and to design mitigations before those models shipped.
Responsibility for that work has been split along threat categories. Bio-risk assessment moves into one existing team, cyber-risk into another, and the rest of the preparedness remit is being distributed across specialty groups already inside OpenAI. There is no longer a single team whose sole job is to evaluate whether a new model is safe to release.
Dylan Scandinaro, who ran the preparedness team after being hired from Anthropic in February 2026, will now focus on the implications of recursive self-improving AI — systems that can rewrite and upgrade themselves. That is a narrower and more research-flavored mandate than the pre-deployment risk review the preparedness team previously owned.
Key facts
- 01OpenAI disbanded its preparedness team at the end of July 2026, according to the Financial Times.
- 02Preparedness duties have been redistributed into existing bio, cyber, and specialty teams rather than a single risk group.
- 03Team head Dylan Scandinaro, poached from Anthropic in February 2026, will now focus on recursive self-improving AI.
- 04Ethics lead Chloé Bakalar, Chief Futurist Josh Achiam, and head of safety Johannes Heidecke have all recently left OpenAI.
- 05The reorganization lands as OpenAI moves toward an expected IPO.
The dissolution follows a run of senior safety and ethics departures. Ethics lead Chloé Bakalar, Chief Futurist Josh Achiam, and head of safety Johannes Heidecke have all left OpenAI recently. Over the past few years the company has also dismantled its AGI readiness team and its superalignment team, the latter of which was originally pitched as a multi-year, compute-backed effort to align systems smarter than humans.
Jan Leike, who co-led superalignment before resigning from OpenAI in 2024 and moving to Anthropic, told the Financial Times that the company is prioritizing what he called
Leike's phrasing echoes the argument he made on his way out in 2024, when he said safety culture and processes had taken a back seat to product velocity. The pattern he described — safety functions absorbed into product teams, senior safety staff exiting, standalone risk groups shuttered — has continued, and the preparedness dissolution is the clearest structural version of it so far.
OpenAI's counterargument, based on the reorganization itself, is that embedding risk work inside product teams gets safety closer to the models being shipped rather than isolating it in a parallel org. That is a defensible operating theory: siloed safety teams can end up flagging risks after launch decisions are effectively locked in. Whether the embedded model actually produces harder pushback on risky launches, or softer pushback because the reviewers now report up through product, is the open question.
The timing matters because OpenAI is heading into an IPO where governance and risk disclosures will be scrutinized by underwriters, regulators, and public-market investors. A company that has serially dismantled its named safety teams — AGI readiness, superalignment, and now preparedness — will need a clean answer for how catastrophic-risk review works in the new structure, who signs off on frontier model releases, and what the escalation path looks like when a bio or cyber team flags something the product org would rather ship anyway.
For the broader AI market, OpenAI's move sets a template competitors will read closely. If the largest lab in the industry decides dedicated preparedness teams are organizationally redundant, other labs will face pressure — from boards and from cost — to consolidate their own safety functions the same way. Anthropic, which lost Scandinaro to OpenAI in February and has publicly leaned into safety branding, is now the clearest counter-position in the market. The next twelve months will show whether preparedness work done inside product teams catches the risks a dedicated group would have, or whether the industry re-learns why the standalone teams existed in the first place.
Working on something we should cover, or seeing a story we missed? Send leads, documents, or feedback to hello@aichatdaily.com. For sensitive tips, see our secure tips page for Signal and PGP options.
Spotted an error? Email hello@aichatdaily.com with the URL and the issue, or read our full corrections policy.
