OpenAI Reportedly Disbands Team Dedicated to Assessing Major AI Risks

OpenAI reportedly disbanded its internal "preparedness team" at the end of last month, according to a report published by the Financial Times and picked up by The Verge. This unit was tasked with assessing whether the company's AI models pose significant risks and with proposing concrete measures to reduce such dangers.

What the "preparedness" team did

According to The Verge, the team's main job was to evaluate extreme scenarios involving the misuse of AI systems. Cited examples include the possibility that a model could be exploited for sophisticated cyberattacks or even behave unpredictably, slipping beyond the control of its developers. These types of risks are considered among the most sensitive topics in the public debate over AI safety.

Responsibilities redistributed across teams

According to the report, OpenAI has not abandoned risk monitoring altogether but has instead decided to divide these responsibilities among several teams specializing in distinct areas, such as cybersecurity and biological threats. In practice, instead of relying on a single, centralized unit, the company is now adopting a fragmented approach, with each risk area managed separately.

Context and possible implications

The decision comes at a time when AI companies are facing mounting pressure from both regulators and the public to demonstrate that they are developing safe technologies. Disbanding a team dedicated exclusively to assessing major risks could raise questions about how OpenAI intends to manage the long-term challenges tied to the safety of its advanced systems.

So far, the company has not offered a detailed public comment on the exact reasons behind this restructuring, and the available information comes solely from the Financial Times report, later picked up by The Verge. It remains to be seen whether this new approach, based on domain-specific teams, will effectively cover all the risk scenarios that the original "preparedness" team was meant to monitor.