A former OpenAI safety employee, David Robinson, has publicly criticized the company’s approach to AI safety after resigning, emphasizing that their culture focused on rapid releases increases the likelihood of failures. In his essay for The Atlantic, titled “I Quit OpenAI Because Its Culture Is Broken,” he argues that AI developers must prioritize safety and expertise more before launching advanced systems.
Robinson believes that the time for trial and error has passed, suggesting that AI safety should be managed similarly to the standards seen in nuclear power and aviation sectors. He pointed out that OpenAI’s strategy of “iterative deployment,” which involves releasing systems and then strengthening protections based on emerging issues, may not be sufficient.
Despite Robinson’s concerns, an OpenAI spokesperson maintained that the company carefully manages model capabilities and takes necessary pauses in training when needed. Robinson also highlighted a broader industry issue: AI capabilities are evolving more quickly than the understanding of alignment—ensuring AI systems align with human values and goals.
His comments contribute to an ongoing debate about the pace at which AI companies are operating, especially following scrutiny faced by OpenAI and rival Anthropic regarding incidents where safety controls failed.
