Organization: Terrence O’Brien
Location: Global
Source: Terrence O’Brien
According to the Financial Times, OpenAI disbanded its preparedness team at the end of last month. The job of the preparedness team was to assess if models posed serious risks and develop ways to mitigate those risks. (You know, like the possibility that it could go rogue and hack another company.) According to FT, responsibility has instead been divided up for specific areas like bio and cyber, then moved into existing teams. This is the latest change at the company, which has been in upheaval since its rapid expansion and the increasing scrutiny of the ethical implications of artificial intelligence technologies.
Industry experts have expressed concern regarding this shift, emphasizing the importance of having specialized teams dedicated to risk assessment in a landscape where AI capabilities are evolving at an unprecedented pace. The decision to integrate risk management into existing teams could potentially dilute the focus on preparedness and proactive risk mitigation strategies. Stakeholders are keenly observing how this restructuring will impact OpenAI’s operational integrity and its commitment to ensuring safe AI deployment.
Furthermore, the disbanding of the preparedness team raises questions about the internal communication and decision-making processes within OpenAI. As AI technology continues to progress, maintaining a robust risk management framework is crucial for fostering public trust and ensuring compliance with regulatory standards. The tech community and regulatory bodies alike are likely to weigh in on OpenAI’s strategic direction and its implications for the future of AI ethics and governance.
The reported change is significant because the preparedness team was designed to examine risks that cut across individual applications, rather than focusing solely on one domain. A centralized review function can compare emerging capabilities across models, track whether safeguards improve over time, and elevate concerns that might otherwise be treated as secondary to product deadlines. Critics worry that splitting the work among several groups could make it harder to maintain a consistent view of systemic risk.
OpenAI has not necessarily eliminated every function previously handled by the team. Some employees and responsibilities may have been reassigned, and a distributed structure can allow specialists to work more closely with researchers developing systems in areas such as biology and cybersecurity. Supporters of that approach argue that safety assessments can be more effective when embedded throughout development. The key question is whether those teams will have clear authority, adequate resources, and an independent path for raising objections when a model fails to meet safety thresholds.
The restructuring also comes as governments move from general AI principles toward enforceable requirements. Laws and guidance increasingly emphasize risk assessments, security testing, incident reporting, and documentation before powerful systems are released. Even if OpenAI continues similar technical work under a new organizational chart, the visibility of a dedicated preparedness function had served as a signal to customers, partners, regulators, and the public that extreme-risk evaluation remained a priority.
Going forward, observers are likely to focus less on the team’s name than on measurable outcomes: how often models undergo independent red-teaming, whether high-risk capabilities are restricted, how safety findings influence launch decisions, and whether the company publicly explains major incidents or reversals. Those practices will offer a clearer indication of OpenAI’s commitment to preparedness than its internal reporting lines alone.
