OpenAI dismantles team tracking catastrophic AI risks
OpenAI has quietly disbanded the unit tasked with spotting existential threats from its own AI models. The Preparedness team, which assessed catastrophic risks from advanced AI systems, is no more—its responsibilities parcelled out to existing groups inside the company. Several safety researchers have already left the organization, according to sources who describe a growing unease within OpenAI about whether the oversight it once had is now fading.
What the Preparedness team did
The Preparedness team focused on evaluating whether OpenAI’s latest AI models could cause severe harm, from large-scale cyberattacks to bioweapon misuse. Its work was meant to provide an independent check on systems that could outpace human control or oversight. Now that work is being absorbed into broader safety and compliance units without a dedicated structure, raising questions about continuity and depth of scrutiny.
Why voices inside are worried
Multiple insiders told The Decoder of a “burbling sense of responsibility and dread” among remaining staff. Some fear that by dissolving the team, OpenAI may be deprioritizing the kind of rigorous, specialized risk assessment that once set it apart in the industry. The departures of safety researchers add to the sense that institutional commitment to catastrophic-risk prevention is waning, even as the company accelerates model releases.
Why it matters
OpenAI’s move strips the company of a dedicated layer of safeguards at a time when AI systems are growing more powerful and harder to predict. The reassignment of risk work doesn’t necessarily mean oversight disappears, but it removes a focused, expert-driven process designed specifically for extreme scenarios. For an industry already scrutinized over safety lapses, this shift signals a realignment of priorities—and it leaves stakeholders to wonder who, exactly, is now watching the watchers.
Source: The Decoder. AI-assisted editorial synthesis — TechnoExpress.