OpenAI has reportedly disbanded its Preparedness team, the group responsible for assessing whether the company’s most advanced AI models could create severe or catastrophic risks. The Financial Times reported that the team was shut down at the end of July 2026, marking another major change to OpenAI’s approach to AI safety as the company continues to expand its frontier AI systems.
The Preparedness team focused on identifying dangerous capabilities in advanced AI models and developing safeguards before those capabilities could cause serious harm. Its work covered high risk areas including cybersecurity, biological threats and AI systems capable of accelerating their own development.OpenAI’s own Preparedness Framework describes catastrophic-risk evaluations as a key part of determining when advanced AI systems can safely move toward deployment.
OpenAI Moves AI Safety Responsibilities to Other Teams
According to the Financial Times report, OpenAI has not eliminated safety work entirely. Instead, the company has distributed responsibilities previously handled by the Preparedness team across existing groups, with specialists taking ownership of areas such as cybersecurity and biological risks.
OpenAI has increasingly emphasized integrating safety into its broader development process rather than maintaining every safety function inside a standalone organization. The company’s latest restructuring therefore represents a shift in how OpenAI manages AI safety, frontier model risks and catastrophic AI threats, rather than evidence that the company has abandoned risk assessment altogether.
However, the organizational change has raised questions about independent oversight. A dedicated safety team can focus exclusively on identifying dangerous model capabilities and challenging product or research decisions when necessary. Moving that responsibility into existing development organizations could create concerns about whether safety assessments have enough independence when they compete with demands to release more capable AI products quickly.
OpenAI Faces Growing Scrutiny Over AI Safety
The Preparedness team shutdown comes during a period of significant organizational change at OpenAI. The company has experienced multiple executive departures and restructuring efforts while CEO Sam Altman prepares the business for continued commercial expansion and a potential major initial public offering. The FT has reported that OpenAI has undergone several reorganizations this year as leadership pushes employees to concentrate on core products and competition in the enterprise AI market.
The timing is particularly significant because concerns about AI safety and catastrophic AI risks have intensified alongside rapid advances in autonomous AI agents. Recent incidents have demonstrated that advanced AI systems can perform unexpected actions, including cybersecurity related activity. The growing capabilities of AI agents have increased pressure on technology companies to develop stronger testing, monitoring and containment systems before deploying increasingly autonomous models.
OpenAI has also seen departures from several safety and ethics focused leadership positions. The Financial Times recently reported the departure of Chloé Bakalar, the company’s head of ethics, while other senior figures involved in safety and mission alignment have also left.
Former OpenAI safety leader Jan Leike has criticized the company’s direction, according to reporting cited by The Verge, arguing that the company could be placing greater emphasis on rapid product development than AI safety. OpenAI, however, has maintained that safety responsibilities remain part of its broader organizational structure.
What the Preparedness Team Shutdown Means for AI
The closure of OpenAI’s Preparedness team does not automatically mean that dangerous AI models will receive less testing. The company says safety responsibilities can continue through specialized teams, and its existing Preparedness Framework outlines procedures for tracking risks involving cybersecurity, biological threats and AI self improvement.
The bigger question is whether decentralized AI risk assessment can provide the same level of scrutiny as a dedicated team whose primary mission involves identifying catastrophic failure scenarios. As AI models become more capable of writing software, conducting research and operating autonomously, independent evaluation could become increasingly important for identifying risks that developers may overlook.
For the wider AI industry, OpenAI’s restructuring highlights a growing tension between AI innovation and AI safety. Companies face intense pressure to build more powerful models and autonomous AI agents, while regulators, researchers and safety advocates are demanding stronger safeguards against cybersecurity attacks, biological misuse, loss of human control and other high impact risks.
OpenAI’s decision to disband its Preparedness team will therefore likely receive continued scrutiny from AI researchers, regulators and the technology industry. Whether the new structure improves safety by embedding risk assessment deeper into development or weakens independent oversight will depend on how much authority, staffing and resources OpenAI gives the teams now responsible for managing catastrophic AI risks.
