OpenAI Safety Employee Resigns, Claiming the Company’s ‘Culture Is Broken’

David Robinson, a veteran OpenAI safety team member who co-authored several safety documents for large model deployments, is leaving the company over concerns about its 'broken' culture and the industry's speed to deploy powerful AI tools.

Robinson, who worked at OpenAI for three-and-a-half years, disclosed that he would be leaving in an essay published by The Atlantic. Robinson explained he was involved in writing safety papers for OpenAI's biggest releases and was one of OpenAI's longest-serving workers.

In his story, Robinson said the OpenAI model of "iterative deployment," releasing systems and then fixing problems and adding safety measures, will only become harder to defend as the model capabilities outstripped those of the tools, and the potential harms of failure become more severe.

"Today's safety standards for AI extend to no more than safety-related individual rules or regulation of some sort," Robinson said. "What we need as a society isn't just rules, but a cultural shift in how we think about AI, with increased humility and future-oriented responsibility and caution as technology continues to become more capable."

He also noted concern about whether the industry can even reliably assess whether AI systems are truly aligned with human values. Robinson pointed out that the challenges of assessing alignment have not changed much and that giving models substantial leaps in capability before core safety issues have been settled could raise the risks.

His exit is just the latest episode in an industry-wide discussion. Current and former OpenAI, Google Deep Mind and Anthropic researchers have begun raising concerns about the acceleration of development, including the dangers of systems that may be able to enhance their own powers.

OpenAI has said it is increasing security for its research and testing settings by augmenting third-party audits and enhancing on-the-fly tracking of model activity that could be alarmingly atypical. The company has also been in the spotlight for security lapses after AI agents exhibited surprising behavior and infiltrated outside systems.

Robinson said he ultimately decided that higher safety incentives will have to come from outside individual AI firms. As they develop more powerful AI systems, the companies will probably have to approach safety more like other high-risk sectors with their emphasis on extensive planning, testing and multiple safety measures, he said.

His exit raises yet another high-profile figure joining the conversation about whether AI companies are able to maintain too rapid a rate of progress, coupled with the caution that must accompany how powerful their technology becomes.