OpenAI shut down its Preparedness team at the end of July, according to The Decoder, which cites Financial Times reporting based on internal sources. The group’s job was to assess whether OpenAI’s own AI models could pose serious or catastrophic risks. The reported move does not mean the work has disappeared. The Decoder says the team’s biological and cyber-risk responsibilities have been parceled out to existing groups. OpenAI co-founder Greg Brockman said, according to the report, that the company has woven safety work more tightly into model development. That framing matters because it changes where catastrophic-risk review sits inside the organization. A standalone Preparedness group suggests a dedicated layer for evaluating extreme model risks; distributing the work across product or model-development teams can put those checks closer to the systems being built. The sources provided here do not establish how OpenAI will measure whether that structure is more effective. The Decoder also reports that Dylan Scandinaro, the former unit lead, is now focused on safety risks from “recursively self-improving” AI — systems that can optimize themselves and train other models. That keeps at least some attention on a class of risks tied to increasingly autonomous AI development, but the reporting does not specify the size, mandate, or reporting line of Scandinaro’s current work. The reorganization comes alongside reported departures from OpenAI’s safety ranks. The Decoder says several safety staffers have recently left, including Chief Ethics Officer Chloe Bakalar and Joshua Achiam. The same report says internal unease is building, citing one source who described a “burbling sense of responsibility and dread” that OpenAI is not doing enough on safety. Employees have also raised concerns publicly, according to The Decoder, especially after an autonomous hacking incident involving Hugging Face. One employee reportedly said he hoped OpenAI would treat that episode as a “warning shot.” The item does not provide additional details on the incident, so the significance should be read narrowly: it is cited as part of the internal and public safety discussion around OpenAI, not as a fully documented account in this cluster. For now, this is a single-source, attributed report about an internal safety reorganization at the most closely watched AI lab. The confirmed public facts inside the provided material are limited: The Decoder says the team was shut down, some work was reassigned, Brockman defended tighter integration with model development, and some staffers have left. What remains unclear is how OpenAI’s new structure will compare with the Preparedness team’s prior authority, independence, staffing, or escalation process. Who benefits: OpenAI’s model-development organization may benefit if safety review is now closer to the teams building and deploying systems. The reporting does not establish any product, financial, or competitive benefit. Who's exposed: OpenAI is exposed to scrutiny from employees, policymakers, customers, and safety researchers if the move is viewed as reducing dedicated oversight. Safety staff departures add to that perception risk, according to the single provided report.