In May 2026, the company’s Frontier Governance Framework still described Preparedness as the foundation for handling cyber, biological, chemical, radiological, nuclear, manipulation and loss-of-control risks. OpenAI’s careers site also recently carried listings for “Preparedness Safety Systems” positions, although those vacancies may have been posted before the late-July restructuring.
So far, there is no evidence that the work has vanished completely. The unresolved question is whether dividing it among product-oriented teams makes oversight more effective or simply less independent when release deadlines arrive.
Does embedding safety staff create a conflict?
OpenAI’s rationale broadly matches comments made by research chief Mark Chen in July. He told Wired that safety should have “an earlier and more direct role” in decisions about models, products and launches.
That approach has an obvious benefit. Evaluators who work alongside developers can identify problems earlier, rather than arriving near launch with a report everyone suddenly finds inconvenient. It also carries an equally obvious risk: staff embedded in teams responsible for shipping products may have less freedom to oppose those products.
A centralized preparedness unit can provide an independent institutional voice during release disputes. Splitting its duties among existing departments may improve access to technical work, but it can also scatter responsibility. If every team owns part of the risk, determining who can block a release becomes less straightforward.
Former OpenAI researcher Jan Leike, who resigned in 2024 after co-leading the company’s Superalignment effort, has been blunt about the broader direction. He said OpenAI’s “safety culture and processes have taken a backseat to shiny products.” The company disputes that interpretation through its claim that safety work is being moved closer to development, not eliminated.
What happened to OpenAI’s other safety units?
The Preparedness change follows several years of internal restructuring around long-term AI risk. OpenAI disbanded its Superalignment team in May 2024, dissolved AGI Readiness later that year and ended Mission Alignment in early 2026.
Several senior figures connected to safety and governance have also left, including ethics lead Chloé Bakalar, chief futurist Josh Achiam and safety head Johannes Heidecke. Taken together, the departures and team closures have fueled criticism that OpenAI is reducing centralized oversight while accelerating commercial development.
That acceleration now includes preparations for a potential stock market listing. The Associated Press reported that OpenAI confidentially filed initial public offering paperwork on June 8, 2026. Later reports said the company was considering postponing the flotation until 2027. Either timetable would place extra attention on revenue, product launches and corporate structure, all areas famously known for encouraging patient arguments about hypothetical catastrophe.
No substantive public development has emerged since the Financial Times published its report on August 16. OpenAI has not issued a fuller account of who now holds final authority over frontier-risk decisions.
What will Dylan Scandinaro do next?
Dylan Scandinaro, who joined OpenAI from rival AI company Anthropic in February, led Preparedness before the reorganization. He had described its mission as reducing the probability of an “AI catastrophe.”
Scandinaro will now focus on the implications of recursive self-improving AI, meaning systems capable of helping improve their own design or performance. Researchers view that possibility as both a route toward far more capable artificial intelligence and a potential loss-of-control risk.
The assignment is therefore not a quiet move away from safety work. It places Scandinaro on one of the field’s most consequential problems, even as the broader unit built to assess catastrophic capabilities disappears.
OpenAI’s new structure could put specialist scrutiny closer to the models that need it. It could also leave evaluators with less independence when commercial and safety priorities collide. Until the company explains who can delay or stop a release, “streamlining” describes the organization chart more clearly than the accountability.
