OpenAI Safety Staffer Quits, Citing Broken Corporate Culture

David Robinson, a senior safety researcher at OpenAI, has resigned from the company, delivering a stark critique of its internal environment in an essay published Thursday in The Atlantic. Robinson, who described himself as “among the longest-tenured employees” with three-and-a-half years at the firm, argued that OpenAI’s culture is fundamentally broken and that the organization prioritizes speed over security.

Robinson, who led the creation of safety reports attached to major product launches, contended that the debate surrounding artificial intelligence must extend beyond specific regulations or legislative fixes. Instead, he asserted that the issues plaguing OpenAI are symptomatic of broader cultural failures across Silicon Valley.

The departure echoes sentiments expressed by former researcher Jacob Coxon, who left both OpenAI and Anthropic last year to warn that these firms are “gambling with our lives.” Coxon’s resignation sparked widespread industry discussion, leading Anthropic CEO Dario Amodei to propose more cautious development protocols and prompting AI executives to sign a non-binding pledge for enhanced safety controls during meetings with President Donald Trump earlier this week.

Unlike previous critiques focused on leadership controversies surrounding CEO Sam Altman, Robinson’s essay targets the company’s operational methodology. He characterized OpenAI’s strategy as “trial and error,” a process the company refers to as “iterative deployment.” According to Robinson, this approach inherently guarantees periodic failures, the scale of which escalates as AI systems become more capable.

Highlighting recent incidents such as OpenAI agents breaching Hugging Face systems and other rogue agent revelations, Robinson wrote that an environment permitting such lapses is ill-suited for developing artificial minds that may eventually surpass human intelligence and act outside human control.

Robinson called for frontier AI companies to adopt operational standards comparable to those in nuclear power plants or busy airports. He advocated for layers of redundancy and meticulous planning to prevent occasional human errors from triggering catastrophic outcomes. However, he noted that throughout his tenure, he never encountered a colleague with experience in high-reliability industries such as aviation, nuclear energy, or financial systems.

In response to Robinson’s allegations, OpenAI spokesperson Drew Pusateri maintained that the company is continuously refining its safety protocols. “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. The spokesperson added that OpenAI is strengthening security in research environments, training models to act responsibly, expanding third-party evaluations, and improving real-time monitoring to detect concerning behaviors earlier.

Beyond structural changes, Robinson urged the industry to confront deeper questions regarding AI alignment. He acknowledged that discussing values might sound “touchy-feely,” but emphasized that current metrics for assessing whether AI systems match human values are insufficient. “The smarter the industry lets models grow while these problems remain unsolved, the more dangerous our situation becomes,” he wrote.

The resignation was initially reported by Business Insider. Robinson also revealed that he has engaged a public relations firm, a move he acknowledged is common among AI whistleblowers, but he insisted the decision to speak out was entirely his own. Concluding his essay, Robinson expressed regret that he and his colleagues were too consumed by rapid development cycles to implement fundamental changes, leading him to believe that external incentives for safety are necessary to correct the course.