What happened

The Financial Times reported that OpenAI disbanded its Preparedness team at the end of last month, a group tasked with assessing whether advanced models could pose serious risks and with developing ways to reduce those risks. OpenAI denied that characterization in a statement to The Verge.

The dispute matters because the team’s remit sat close to the center of frontier AI safety. Preparedness work is about looking ahead: testing whether future or current models might create high-impact hazards, such as cybersecurity abuse, biological or chemical risks, or dangerous forms of autonomous behavior.

OpenAI’s response

OpenAI spokesperson Kayla Wood told The Verge: “We have not disbanded the Preparedness team.” She said OpenAI has research leaders across cybersecurity, biological and chemical risks, and AI self-improvement capabilities, all reporting to Saachi Jain, the company’s head of safety.

The Financial Times, however, reported that responsibility has been split across specific domains, such as bio and cyber, and moved into existing teams. That leaves the key question less about whether safety work still exists and more about how independent and visible that work remains inside the company.

Key facts from the report include:

• FT says the team was disbanded at the end of last month.
• OpenAI denies the team was disbanded.
• Safety areas named include cyber, biological and chemical risks, and AI self-improvement.
• Dylan Scandinaro, head of the Preparedness team and hired from Anthropic in February, will focus on the implications of “recursive self-improving” AI.

A broader pattern of change

The reported move follows a series of organizational changes at OpenAI. The company has previously dissolved its AGI readiness and superalignment teams. In simple terms, AGI readiness concerns preparation for broadly capable AI systems, while superalignment refers to research on keeping systems more capable than humans aligned with human goals and constraints.

The Verge report also notes recent departures including ethics lead Chloé Bakalar, chief futurist Josh Achiam, and head of safety Johannes Heidecke. Jan Leike, who left OpenAI in 2024, told the Financial Times that the company was favoring “shiny products” over safety.

These changes come as OpenAI is described as moving toward what is expected to be a massive IPO. In that context, any shift in the structure of safety teams becomes a signal watched by employees, critics, customers, and regulators.

Why structure matters

Embedding safety experts inside product or research teams can make risk review more closely connected to day-to-day development. But it can also raise concerns about independence, especially if outside observers cannot see whether safety findings can delay or alter launches.

The reference to recursive self-improving AI is particularly sensitive. The term describes systems that could improve their own capabilities over successive cycles. The report does not say OpenAI has such a system or disclose any specific product capability, but the topic remains central to long-term AI safety debates.

Outlook

This episode shows that AI safety is now an organizational governance issue as much as a technical one. OpenAI’s denial addresses the narrow claim of disbandment, but the larger question is whether safety work retains authority as the company commercializes and prepares for a major public-market moment. The next phase of industry scrutiny will likely focus on transparency, independent risk assessment, and whether internal safety teams can meaningfully shape release decisions.