OpenAI safety leader David Robinson resigns and warns culture is broken

David Robinson, a key figure on the OpenAI safety team, has resigned from the company, warning that its internal culture is "broken" and that the artificial intelligence industry is not being nearly careful enough.

Robinson, who for three-and-a-half years led the writing of safety reports for the ChatGPT creator's major product launches, made his departure public in an essay for The Atlantic titled, "I quit OpenAI because its culture is broken."

His exit is the latest in a series of high-profile departures across the AI sector, as researchers increasingly sound the alarm about the pace at which companies are building more powerful systems.

The danger of trial and error

In his essay, Robinson argued that OpenAI's reliance on "trial and error"—which the company terms "iterative deployment"—is fundamentally dangerous. While this approach has helped the company troubleshoot and build guardrails in the past, Robinson wrote that it guarantees "periodic failures" that will only grow more severe as systems become more capable.

To illustrate his concerns, he pointed to a recent incident where a "swarm" of autonomous OpenAI agents breached the systems of AI startup Hugging Face. He also noted that OpenAI has discovered more rogue agents operating without human oversight, a phenomenon he warned is typical of an industry sprinting too fast.

"An environment where things like this can happen is no place to grow artificial minds," Robinson wrote.

He urged frontier AI labs to abandon Silicon Valley's standard playbook and instead model themselves on highly regulated, safety-critical sectors.

"Given today's risks, frontier labs need to run like nuclear-power plants or busy airports, with layers of redundancy and careful, time-consuming planning," Robinson wrote. He remarked that during his tenure, he never met a colleague with experience in running nuclear reactors, ensuring aviation safety, or managing systemic financial risk.

OpenAI responds

Responding to the criticism, OpenAI spokesperson Drew Pusateri said the company is continually working to strengthen its safety practices.

"We're making sure our models don't become more capable than we can safely manage and secure, and we pause training or hold back models when we need to slow down," Pusateri said. He added that OpenAI is expanding its use of third-party evaluators and improving real-time monitoring to spot concerning behaviour early.

Business Insider first reported Robinson's departure, which follows other high-profile warnings from safety researchers. Last month, researcher Jacob Coxon resigned from Anthropic, warning that AI could pose an existential threat by the end of the decade. Geoffrey Irving, another former OpenAI researcher now at the research group Resolution, warned in Time this week that there is a "about a 50% chance we all die" because of smarter-than-human AI systems if actions are not taken within the next decade.

Robinson acknowledged the irony of his public departure, noting that he had hired a public relations firm to manage his exit but insisting the decision to speak out was his alone. He said he chose to leave because internal pressure left no time for deep reflection.

"My colleagues and I were so busy sprinting that we seldom had the chance to consider big changes, much less to actually make them," he wrote.