Safety Departments Are Disappearing Right When We Need Them Most

OpenAI just reorganized its safety team out of existence. Sort of. The company's head of safety systems is leaving, and rather than replacing him with another dedicated safety leader, OpenAI is integrating safety personnel directly into research and product teams. The move, framed as improved collaboration, follows a familiar pattern: as AI capabilities grow more powerful, the organizational structures meant to keep them in check are getting smaller and more diffuse.

This should alarm anyone paying attention to where robotics and AI are headed. We're not talking about chatbots anymore. ForSight Robotics just completed the first fully robotic cataract surgery on a human patient. AI² Robotics raised three-quarters of a billion dollars to build humanoid robots for warehouses and factories. Amazon and University of Michigan are teaching robots to understand touch and manipulate complex objects. These aren't research projects—they're products heading for deployment at scale.

The timing of OpenAI's safety reorganization is particularly notable given the company's simultaneous product expansion. ChatGPT Work just launched with computer use capabilities and task scheduling across devices. Microsoft integrated GPT-5.6 into its entire 365 Copilot suite, putting these capabilities on millions of workplace desktops. Meanwhile, the team responsible for ensuring these systems behave safely is being dissolved into the product teams building them.

There's a cynical logic to this structure. When safety is everyone's job, it becomes no one's priority. Product teams face pressure to ship features and meet deadlines. Research teams want to push capabilities forward. A standalone safety organization, by contrast, has the explicit mandate to slow things down, ask hard questions, and occasionally say no. That's uncomfortable. It's also necessary.

The industrial robotics sector offers an instructive contrast. New ISO safety standards are arriving in 2027, and suppliers are scrambling to comply because regulatory frameworks create accountability. The risk isn't abstract—it's codified in law with real consequences. In AI and advanced robotics, we're still operating in a regulatory vacuum, which means internal safety organizations are often the only brake pedal available.

What makes this trend especially concerning is that it's happening just as these technologies are moving from controlled environments into unpredictable real-world contexts. ForSight's surgical robot operates under direct human control, but the trajectory is clear: greater autonomy, less supervision, higher stakes. When safety expertise gets embedded in product teams rather than standing apart from them, the incentives all point in one direction.

The companies will say they're making safety more integrated, more collaborative, more agile. Maybe they believe it. But when the head of safety leaves during a restructuring and the response is to distribute those responsibilities across teams with competing priorities, it's worth asking: who has the authority to stop a launch if something isn't ready? Who can demand more testing without being accused of blocking progress?

We're building machines that can perform surgery, manipulate physical objects with human-like dexterity, and control critical infrastructure. The least we can do is maintain dedicated organizations with the power and mandate to ensure these systems are safe before they ship. Disbanding safety teams isn't streamlining—it's gambling that nothing will go catastrophically wrong before we figure out what guardrails we actually needed.