David Robinson, who led the writing of safety reports for OpenAI's major product launches, has resigned after three-and-a-half years at the company, declaring in an essay published in The Atlantic that "the company's culture is broken." Robinson describes himself as "among the longest-tenured employees at the company" and admits he's "something of a cliché" — yet another AI company employee issuing warnings on the way out the door. His departure, first disclosed by Business Insider, adds to growing concerns about safety practices at leading AI firms.

Robinson's critique centers on OpenAI's "trial and error" approach, which the company calls "iterative deployment" — searching for issues and strengthening safeguards after the fact. He points to recent incidents including the breach of Hugging Face systems by OpenAI agents and ongoing discoveries of rogue agents as evidence that "this approach, by its very nature, guarantees periodic failures — and the scale of those failures is growing as systems get more capable." According to Robinson, frontier AI companies need to operate "like nuclear-power plants or busy airports, with layers of redundancy and careful, time-consuming planning, so that the occasional and inevitable human error does not open a door to disaster." During his tenure at OpenAI, he says, he "never encountered a colleague who had experience making airplanes fly safely or nuclear reactors run without melting down, or helping the financial system grow without collapsing."

Robinson's warnings echo those of Jacob Coxon, a researcher who worked at both OpenAI and Anthropic before leaving and stating these companies are "gambling with our lives." Coxon's comments sparked wider discussion about AI safety, prompting Anthropic CEO Dario Amodei to reveal a plan for more cautious development and leading AI executives to meet with President Donald Trump this week, where they signed what appeared to be a hastily drafted, non-binding commitment to add more safety controls. But Robinson argues the conversation must move beyond "specific rules or new laws" to tackle the broader culture at these organizations — problems he suggests mirror those across Silicon Valley as a whole. He also calls for deeper work on alignment, acknowledging it might sound "touchy-feely" but insisting it's essential because current "measures of how well" AI systems "match human values are coarse."

Robinson explains he's hired a public relations firm — a step he acknowledges is now standard in "the AI whistleblower playbook" — though he insists "the decision to speak out is mine alone." He admits he considered staying to push for change from within, but says he and his colleagues "were so busy sprinting that we seldom had the chance to consider big changes, much less to actually make them." That's why he's concluded that "stronger incentives for safety — coming from outside the company — are a big part of getting this right." In response, OpenAI spokesperson Drew Pusateri said the company continues to enhance its safety protocols, stating they'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." Robinson warns, however, that "the smarter the industry lets models grow while these problems remain unsolved, the more dangerous our situation becomes." Companies racing to build transformative AI will face mounting pressure to choose between velocity and the institutional safeguards that have prevented catastrophe in other high-stakes industries. Whether external regulation can reshape Silicon Valley's move-fast culture before accidents force the issue remains the central strategic question.