By his own admission, David Robinson is “an ordinary person.” An employee at a major AI company issued a serious warning as he left the company.
In an essay published in The Atlantic, Robinson said he led the creation of safety reports accompanying OpenAI’s major product launches. He also said he has been with OpenAI for three and a half years, making him “one of the company’s longest-tenured employees.” He is now about to quit the company. Because, in his view, the company’s “culture is broken.”
In some ways, Robinson’s comments echo those of Jacob Coxon, who worked as a researcher at both OpenAI and Anthropic before leaving the company and declared that these companies were “risking our lives.” Coxon’s comments led to a broader discussion about the safety of AI, and Anthropic CEO Dario Amodei revealed plans for more cautious AI development. AI executives met with President Donald Trump this week and signed a non-binding pledge that appeared to be hastily written to implement additional security controls.
But in Robinson’s view, the discussion needs to go beyond “specific rules and new laws” and address the overall culture of these companies. And while much of the reporting on OpenAI has focused on how CEO Sam Altman lost the trust of his former colleagues, Robinson’s essay suggests that OpenAI’s culture problems are the same as those in Silicon Valley as a whole.
“OpenAI has grown through trial and error (what the company calls ‘iterative deployment’) of looking for problems and improving guardrails in response,” he wrote. “But by its very nature, this approach guarantees periodic failures, and as the system’s capabilities improve, the magnitude of those failures grows.”
Pointing to the recent breach of the Hugging Face system by OpenAI agents and continued revelations that OpenAI is discovering more rogue agents, Robinson argued that “an environment where something like this could happen is not the place to grow an artificial mind that is smarter than us and may not do what we want.”
Given the increased risk, Robinson argued that frontier AI companies need to begin operations with “layers of redundancy and careful, time-consuming planning to ensure that the occasional, inevitable human error does not open the door to disaster.”
But during his time at OpenAI, Robinson said, “I never met a colleague who had the experience of flying an airplane safely, running a nuclear reactor without meltdown, or helping a financial system grow without collapsing.”
In response to Robinson’s essay, OpenAI spokesperson Drew Pusateri said the company continues to improve its safety measures.
“We are careful not to exceed the capacity of our models to safely manage and secure them, and we pause training or throttle models when we need to slow them down,” Pusateri said in a statement. “We are making significant changes to strengthen the security of our research and test environments, train our models to not only complete tasks but also perform them responsibly, expand our work with third-party evaluators, and improve real-time monitoring to detect and respond to concerning behavior early in the training process.”
In addition to calling for a change in OpenAI’s culture, Robinson also said it’s time to ask bigger questions about its integrity. He acknowledged that this may sound “sensational”, but said alignment is important because companies currently have “crude standards” for determining how well their AI systems align with human values.
“The longer the industry allows the model to grow without these issues being resolved, the more dangerous our situation becomes,” he says.
Robinson’s departure was first reported by Business Insider. In his essay, he also acknowledged following apparently common steps in an AI whistleblower strategy. That means they hired a PR firm. However, he insisted that “the decision to speak out or not is solely mine.”
“Perhaps I should have stayed here and fought for fundamental changes in staffing and culture, but the reality is that my colleagues and I were so busy sprinting that we had little opportunity to consider, much less actually make, major changes,” Robinson said. “That’s why we concluded that stronger external safety incentives are a big part of getting this right.”
If you buy through links in our articles, we may earn a small commission. This does not affect editorial independence.
