Is OpenAI changing its focus on safety?

OpenAI has disbanded its preparedness team and shifted its responsibilities into existing specialized units that focus on specific areas like cyber security and biological risks. This follows a string of departures of high-level researchers and managers who previously focused on ethical oversight and long-term safety. Critics of the move suggest that the company is shifting its internal priorities away from deep, preemptive safety research to focus more heavily on releasing new products. The leader of the former preparedness team will now dedicate his time to studying the specific risks of AI that can rewrite its own code to improve itself.

Think of an AI model like a powerful tool being built by a large team. When a company has a dedicated safety team, those people act like an independent inspector at a car factory. Their only job is to try to break the car or find hidden flaws before it leaves the plant. By folding these inspectors into other teams, the company is changing the structure of how that safety work gets done. Instead of a separate, centralized team asking big, existential questions, individual project teams are now responsible for spotting the risks within their own specific tasks. This approach makes safety a part of the everyday building process, but it also removes the wall that used to exist between the people who push for new features and the people whose sole job was to stop and check for danger.

As these companies prepare to become public businesses, they face a classic conflict between the desire to launch popular, advanced products and the need to proactively hunt for rare, catastrophic failures. While distributing safety responsibility across every team might make safety a daily habit for all engineers, it also removes the independent voices that were once dedicated to slowing things down. For the public, this change makes it harder to know whether the guardrails on future technology are getting stronger or just becoming a lower priority in the race to build the next big thing.