In a stunning turn of events that proves Silicon Valley’s understanding of irony remains firmly in beta, OpenAI has terminated employees for the cardinal sin of sharing information with an outside AI evaluation group. The crime? Collaborating with researchers who were, checks notes, trying to make AI systems safer.
The company’s investigation concluded that these workers had mishandled “sensitive information”—a phrase that apparently now includes basic research findings that might help the world understand whether the technology they’re building will eventually decide humanity is inefficient and needs optimizing away.
What makes this particularly rich is the theatrical horror with which OpenAI treated the breach. These were not employees selling proprietary algorithms to competitors or posting API keys to Reddit. They were talking to safety researchers. The kind of people whose entire job is to ask uncomfortable questions like “but what if it goes wrong?” The very conversations you’d think a responsible AI company would encourage over catered lunches and ping-pong tables.
The irony is so dense it requires its own computational overhead: a company built on the premise of democratizing AI safety has decided that safety research conducted by external experts constitutes a firing offense. It is as if a pharmaceutical company fired researchers for discussing drug trial results with the FDA.
The real sensitive information here is not what was shared—it is what was revealed about how these companies actually think about accountability. When your biggest fear is not that your AI system might cause harm, but that someone might talk about whether it could, you have discovered something important about your own priorities. And it is not the kind of discovery that makes for good PR.