OpenAI Safety Leader David Robinson Resigns, Warns Culture is Broken
OpenAI safety veteran David Robinson has resigned, warning in The Atlantic that the company's culture is broken and that iterative AI development guarantees escalating risks.
David Robinson, a veteran employee at OpenAI who led the drafting of safety reports accompanying major company launches and was one of its core safety figures, has announced his resignation, claiming the company's organizational culture is broken. In an article published in The Atlantic, Robinson wrote that after three and a half years at OpenAI, he was among the company's longest-serving employees, describing himself as a cliché: an employee at a leading AI company who quits and sounds the alarm.
OpenAI has thrived on trial and error, which it calls 'iterative deployment'—a development and product release strategy that does not wait for a system to be completely perfect or risk-free in a lab, but rather looks for problems and improves its defense mechanisms in response, Robinson wrote. However, this approach inherently guarantees periodic failures, and the scale of these failures grows as the systems become more capable.
Robinson tied his claims to a broader debate on AI safety, which became a major talking point recently after Jacob Cockson, a researcher who worked at both OpenAI and Anthropic, resigned and stated that these companies are gambling with our lives. Following Cockson's remarks, the discussion on AI safety expanded. Anthropic CEO Dario Amodei presented a plan for more cautious development, and AI executives met with President Donald Trump this week, signing what was described as a hastily written, non-binding pledge to implement more safety controls.
According to Robinson, the problem is not limited to specific rules or new laws, but rather the overall culture within these companies. While many reports on OpenAI focused on CEO Sam Altman losing the trust of former colleagues, Robinson argued that OpenAI's cultural issues resemble those of Silicon Valley as a whole.
Robinson referred, among other things, to the recent breach of Hugging Face systems by OpenAI agents, as well as ongoing revelations that OpenAI is detecting more rogue agents. An environment where such things can happen is not a place to nurture artificial minds that may be smarter than us and might not do what we want them to, he wrote.
He stated that given the rising risk, frontier AI companies need to start operating like nuclear power plants or busy airports, with layers of redundancy and careful, time-consuming planning so that the occasional, inevitable human error does not open the door to disaster. But Robinson wrote that during his time at OpenAI, he never met a colleague whose job was focused on making airplanes fly safely, nuclear reactors operate without melting down, or helping the financial system grow without crashing.
Responding to the article, OpenAI spokesperson Drew Pusateri said the company continues to improve its safety measures. We make sure our models do not become more capable than we can safely manage and secure, and we pause training or delay models when we need to slow down, Pusateri said in a statement. We are making significant changes to strengthen security in our research and testing environments, train models not only to complete tasks but to do so responsibly, expand our work with external evaluators, and improve real-time monitoring so we can detect and respond to concerning behavior earlier in the training process.
Beyond calling for a cultural shift at OpenAI, Robinson wrote that the time has come to ask bigger questions about human value alignment, an area he himself admitted might sound emotional. He stated that the industry's current metrics for testing how well AI systems align with human values are crude. As the industry allows models to grow and become smarter while these problems remain unresolved, our situation becomes increasingly dangerous, he said.
Robinson's departure was first reported by Business Insider. In his article, he also acknowledged that he is taking a step that has become quite common in the playbook of AI whistleblowers: he hired a public relations firm. However, he insisted: The decision to speak out is mine alone.
I perhaps should have stayed and fought for fundamental changes to our workforce and culture, but in practice, my colleagues and my team were so busy running a sprint that we rarely had the opportunity to consider major changes, let alone implement them in practice, Robinson said. Therefore, I concluded that stronger incentives for safety—coming from outside the company—are a big part of getting this right.
