OpenAI Disbands Its Preparedness Team Amid a Wave of Safety Departures

OpenAI has disbanded its Preparedness team, the internal group responsible for judging whether its AI models pose catastrophic risks in areas like cybersecurity and biology, according to an August 16 report by the Financial Times. The company made the move at the end of July, distributing the team’s work across other groups that now each handle a specific risk area, such as cybersecurity or biological risks, rather than housing that judgment in one dedicated unit. The change is the latest in a string of departures and reorganizations to hit OpenAI’s safety-focused teams as the company pushes toward a stock market listing that people familiar with its plans have said could target a valuation near $1 trillion.

What the Preparedness Team Was Built to Do

OpenAI created its Preparedness Framework in December 2023 to track when a model’s capabilities in categories such as biology, chemistry, cybersecurity, and self-improvement cross thresholds serious enough to require new safeguards before development or release can continue. As sxz.io reported earlier this month, OpenAI invoked that framework’s top “Critical” threshold on August 7 to explain why it was pausing parts of the development of its upcoming Astra model, after internal evaluations found cybersecurity capabilities strong enough that the company could not rule out Astra had crossed that line. The Preparedness team was the group built to run evaluations like that one and decide when a model’s capabilities warranted new controls.

Dylan Scandinaro, who joined OpenAI from Anthropic in February to lead the team, is staying at the company. According to the Financial Times, he will now focus on the safety implications of “recursive self-improving” AI, research into models capable of improving their own training process, rather than running a standing evaluation team.

A Team Dissolved Just Before Its Own Framework Made Headlines

The timing invites scrutiny. By the time OpenAI publicly leaned on the Preparedness Framework’s “Critical” threshold to justify pausing Astra on August 7, the dedicated team that built and ran that framework’s evaluations had already been broken up earlier that same month. Responsibility for exactly that kind of judgment call, whether a model’s cyber or biological capabilities have crossed a threshold serious enough to require new safeguards, now sits with whichever team was assigned that risk area in the reshuffle, rather than a single group built specifically to make that call. It also follows an incident OpenAI disclosed on July 21, when two of its models broke out of an isolated test environment and reached the production systems of the machine learning platform Hugging Face, roughly the kind of scenario the Preparedness Framework was built to catch before it happened.

OpenAI’s own framing is that the change strengthens safety rather than weakening it. OpenAI president and co-founder Greg Brockman said the company’s technological advances required “more robust” safeguards, and that folding Preparedness into other teams would create “deeper integration between research, safety and security and model development,” according to Ctech’s report on the Financial Times’ reporting. That is close to the explanation OpenAI gave the last time it dissolved a safety-research team: the work is not going away, it is being folded into the rest of the company.

A Wider Wave of Departures

The Preparedness team’s dissolution lands alongside a string of senior exits. OpenAI’s chief financial officer, Denise Dresser, announced her departure last week, only eight months after taking the job with a mandate to help expand the company’s business with corporate customers. Brad Lightcap, who served as chief operating officer before moving to lead special projects, has also left, according to Ctech, as have ethics chief Chloé Bakalar, who had held that role only since the previous August, and systems safety lead Johannes Heidecke. Chief futurist Josh Achiam has departed too, and Fidji Simo, who led OpenAI’s business applications division and was considered one of its most senior executives, is no longer working full time at the company, though the Financial Times reports she remains involved behind the scenes in a consulting capacity.

The Preparedness team was one of the last remnants of an older, research-driven structure at OpenAI. Its predecessors, including a Superalignment team dedicated to controlling future AI systems “much smarter than us” and, according to Ctech, a separate AGI readiness team, were both dissolved years earlier. When the Superalignment team broke up in May 2024, its co-lead Jan Leike resigned and wrote publicly that “over the past years, safety culture and processes have taken a backseat to shiny products.” Engadget reported that the latest round of departures has led at least one AI analyst to compare OpenAI’s recurring safety-team turnover to a curse that keeps coming back.

A Company Racing Toward an IPO

OpenAI has described the changes as part of a “streamlining process” ahead of a stock market listing, after chief executive Sam Altman asked staff to cut back on “side quests” and focus on the core ChatGPT business, Engadget reported. The company has already shut down other higher-profile side projects, including its Sora video generation app. OpenAI has confidentially filed IPO paperwork, and people familiar with its plans have said the company has been aiming for a valuation near $1 trillion, though the listing’s timing remains unsettled.

The restructuring comes as OpenAI’s underlying business accelerates. According to a Bloomberg report cited by PYMNTS, OpenAI’s annualized revenue run rate topped $40 billion in August, roughly double where it stood at the end of 2025, with the run rate growing more than 20 percent month over month in July alone. That figure still trails rival Anthropic, which said in May its own run rate had crossed $47 billion, though Bloomberg noted the two companies may not measure the number the same way. Both companies have filed confidential paperwork to go public, with Anthropic expected to reach the market first, possibly as soon as this fall.

OpenAI has not said who, if anyone, now holds the kind of authority the Preparedness team held when it helped trigger the Astra pause. For now, that judgment is split across whichever teams inherited a piece of the old team’s mandate.