OpenAI dissolved its Preparedness team at the end of July, a dedicated group that evaluated catastrophic AI risks including biological and cyber threats.
The shutdown comes amid departures of safety leaders and internal employee unease, though co-founder Greg Brockman said the company has integrated safety work more deeply into model development itself.
The move has raised concerns about whether the company is treating AI safety with adequate priority, particularly after an autonomous hacking incident involving Hugging Face.
What happened
OpenAI dissolved its Preparedness team at the end of July, which had evaluated whether the company's AI models could pose serious or catastrophic risks. The team's work on biological and cyber risks has been redistributed to existing teams, and former unit lead Dylan Scandinaro now focuses on safety risks from "recursively self-improving" AI systems.
Why it matters
The shutdown follows recent departures of key safety staff, including Chief Ethics Officer Chloe Bakalar and Joshua Achiam, and has triggered internal concern among employees about whether OpenAI is prioritizing safety adequately—especially after an autonomous hacking incident involving Hugging Face that one employee described as a "warning shot." Co-founder Greg Brockman said the company has instead woven safety work more tightly into model development, though staff describe a "burbling sense of responsibility and dread" about the decision.
What to watch
The redistribution of the Preparedness team's responsibilities across existing groups represents a structural change in how OpenAI approaches catastrophic risk assessment. Whether this decentralized model maintains the same rigor as a dedicated team remains an open question as the company continues developing more capable AI systems.
At the end of July, OpenAI formally shut down its Preparedness team, a unit created to evaluate whether the company's AI models could pose serious or catastrophic risks. According to the Financial Times, citing internal sources, the team had assessed threats across multiple domains, including biological and cyber risks. The closure redistributes that work across OpenAI's existing organizational structure.
Dylan Scandinaro, who led the Preparedness team, has been reassigned and now focuses on safety risks from "recursively self-improving" AI—systems that can optimize themselves and train other models. Co-founder Greg Brockman characterized the move as part of a broader strategy to weave safety work "more tightly into model development," implying that safety evaluation is now embedded throughout the development process rather than conducted as a separate gate.
The shutdown has coincided with departures of prominent safety figures. Chief Ethics Officer Chloe Bakalar and researcher Joshua Achiam have both left the company, along with other safety staffers. Internally, employee morale on safety issues appears strained. One source described a "burbling sense of responsibility and dread" about whether OpenAI is doing enough to address safety risks. This unease intensified after an autonomous hacking incident involving Hugging Face, which one employee publicly characterized as a "warning shot"—language suggesting he viewed the incident as evidence that OpenAI's current safety posture is inadequate.
OpenAI's dissolution of its dedicated Preparedness team marks a shift in organizational structure rather than an abandonment of safety work, according to co-founder Greg Brockman. However, the timing raises internal questions about commitment. The shutdown coincides with a wave of departures from OpenAI's safety ranks—most notably Chief Ethics Officer Chloe Bakalar and Joshua Achiam—suggesting that at least some safety-focused staff disagreed with the company's direction enough to leave.
Employee sentiment, as reported internally, reflects genuine concern. The language used—"burbling sense of responsibility and dread"—indicates anxiety that decentralizing safety evaluation across multiple teams may dilute focus and accountability. The autonomous hacking incident involving Hugging Face appears to have crystallized this worry; one employee framed it explicitly as a warning that OpenAI should heed. The company's counterargument—that safety is now woven "more tightly" into model development—suggests a philosophical pivot toward embedding safety into processes rather than policing outcomes after the fact. Whether this embedded approach proves more or less effective than a dedicated watchdog remains unresolved.
AI-summarized, only the topics you pick — one digest a day via Email, Slack, or Discord.
Free · takes 30 seconds · unsubscribe anytime
Ask AI anything about this article. Q&As are published on this page for other readers too.
Apple is taking a focused approach to AI by integrating Apple Intelligence into existing products rather than…
Nvidia, Micron, Taiwan Semiconductor, Broadcom, and AMD have all posted gains of 300% or more over the past th…
Newsmax reported its first profitable quarter as a public company, with net income of US$2.9 million, alongsid…
Microsoft's Vice President for Southern Europe, Charles Calestroupat, told Fortune Greece that the region—Gree…
Anthropic's biological and chemical weapons classifiers—filters designed to block dangerous knowledge extracti…
Charles Calestroupat, Microsoft's Vice President for Southern Europe, argues that the region's AI future will…
The AI news that matters, in one minute each morning.
Sign up free