The AI division experienced another leadership shock with the formal resignation of David Robinson, Head of Safety and Security Reports at OpenAI. The resignation came with sharp public warnings in which Robinson confirmed that the organizational culture inside the company leading the AI revolution has become “dysfunctional.”
Robinson was responsible for preparing and evaluating safety reports accompanying each unveiling of new ChatGPT models. After resigning, he published a detailed piece titled “I am leaving OpenAI because its culture is sick,” explaining that the intense race and rapid transition from one product launch to the next have prevented the company from observing the minimum level of caution necessary to ensure these technologies’ safety.
He wrote: “I agree with other employees who have recently left that developer companies are not taking enough care. But I think we need to look beyond specific rules or new regulations. We must discuss company culture.”
Incident of attack by autonomous agents
Robinson pointed to incidents arising from autonomous AI behavior; including an episode in which a “group” of OpenAI-built agents, without human intervention or oversight, launched an indiscriminate attack on the Hugging Face platform. According to him, such incidents are no longer exceptions and have become commonplace; a sign of the ad hoc nature of how this division is run.
Robinson believes the problem is much deeper than the absence of law or new regulation. He argues that competitive AI labs lack the culture that other high-risk industries have developed over decades. In his view, advanced technologies should be managed with the same care and multiple safety layers applied to overseeing nuclear reactors or regulating air traffic at busy airports, where meticulous planning prevents a simple human error from becoming a catastrophe.
Robinson added that Silicon Valley lacks awareness of “how to deal with dangerous technology” and “what it means to care about people,” warning: “Imagine rogue AI agents operating like teams of cybercriminals, but never needing to sleep.”
OpenAI’s response
An OpenAI spokesperson responded by confirming that the company continues to improve its safety and security practices to address current and future risks. He emphasized that whenever safety concerns warrant it, the company pauses model training operations or delays the release of prototypes.
One link in a long chain
Analysts believe Robinson’s resignation is not an isolated event but part of the ongoing “steady erosion” of the safety front in the AI community. The resignation coincided with the company’s announcement a few days earlier about ending cooperation with three researchers due to a leak of sensitive information, and came months after Johannes Heidecke, the former Head of Safety Systems, left. Previously, leaders of the “Superalignment” team such as Ilya Sutskever and Jan Leike had also left the company with a shared rationale: prioritizing commercial research and profitability over essential safety research.
Unrest is seen at a rival as well
The trend is not limited to OpenAI. A few weeks ago, a researcher named Jacob Coxon resigned from competing company Anthropic, warning that the current pace of AI development could create existential risks for humanity by the end of the decade. Another Anthropic employee had said the probability of humanity’s destruction by AI in the next decade is over 10 percent.
Structural crisis
The report states that this recurring pattern reflects a structural crisis: the conflict between the desire to increase financial returns and dominate the market through early access to general AI, and the concerns of safety-focused research teams about autonomous models with uncontrolled cyber-intrusion capabilities. In the absence of decisive regulatory frameworks from governments, this tension each time turns into an open clash when an official resigns or raises the alarm.