David Robinson, OpenAI's safety report writer, quits and warns of broken culture

David Robinson used to write the safety reports that accompanied every major model release at OpenAI. This week, he resigned from that position and is now speaking out in an editorial in The Atlantic, warning that the culture across the AI industry is fundamentally broken. His words carry weight because he was the person tasked with documenting the risks before each launch - the one whose job was to put the guardrails on paper. If he says the system is broken, it is not a casual complaint; it is a diagnosis from the inside of the machine that builds the models shaping the next decade of computing.

Robinson's core argument is that the problem runs deeper than simply slapping a few new rules or regulations on how we handle training models. He is pushing back on the idea that a compliance framework can fix what is essentially a cultural failure. That distinction matters for every board and executive team that has been treating AI safety as a checklist item - a box to tick before release. Robinson is saying that the incentives inside AI labs are misaligned: speed, scale, and competitive pressure routinely override the cautious voices that are supposed to be the last line of defense. When the person writing the safety reports resigns and goes public, it is a signal that those checks are not just being ignored; they are being structurally undermined.

For context, OpenAI has been at the center of the AI safety debate since its founding, positioning itself as the lab that would prioritize safe development over reckless deployment. Yet the company has repeatedly faced criticism from former employees and researchers who argue that commercial pressures have eroded those original commitments. Robinson's resignation adds to that pattern, but his specific role makes it more pointed. He was not a researcher on the fringe; he was the one who produced the official safety narratives that accompanied each model release. His decision to leave and speak out suggests that the internal process he was part of was not functioning as intended - that the reports were perhaps more about public reassurance than genuine risk mitigation.

The broader industry implication is that self-regulation is failing. Robinson's point that rules alone won't fix the culture is a direct challenge to the approach many companies have adopted: hire a safety team, publish a policy, and claim responsibility. But if the people inside those teams are quitting and sounding alarms, the policies are not working. For executives at AI companies, this is a moment to ask hard questions about whether their safety teams have real authority or are just window dressing. Do they have the power to stop a launch? Are their concerns escalated to the board? Or are they overruled in the name of shipping faster than a competitor?

Regulators are watching. Robinson's editorial lands at a time when governments are drafting AI rules, from the EU's AI Act to state-level initiatives in the US. His argument that cultural change is needed, not just regulation, could influence how those rules are written. If regulators believe that internal safety processes are performative, they may push for more external oversight, third-party audits, or even licensing requirements. That would be a significant shift for an industry that has largely been allowed to police itself. For companies that have been relying on self-assessment, this is a strategic risk that cannot be ignored.

For peers in similar roles - safety officers, ethics leads, compliance heads - Robinson's move is both a warning and a template. It shows that speaking out can come at a personal cost, but also that silence has its own risks. The industry's credibility is at stake. Every time a safety employee resigns and goes public, it reinforces the perception that AI labs are not to be trusted with their own oversight. That perception could lead to a talent drain, as the most principled people decide they cannot work in a system that ignores them. It could also lead to a regulatory backlash that affects everyone.

The strategic takeaway for CEOs and boards is that safety cannot be a checkbox. It has to be embedded in the culture, with real incentives for raising concerns and real consequences for ignoring them. Robinson's resignation is not just a story about one person leaving a company; it is a signal that the current approach is failing. The question is whether other leaders will listen before the next safety report writer quits - or before a model release goes wrong in a way that no report could have predicted.