David Robinson worked at OpenAI for about three and a half years, during which time he was responsible for writing security reports accompanying the release of major models. He stated that he resigned because the company culture "had become dysfunctional." In an article published in The Atlantic Monthly, Robinson mentioned that he oversaw the security reports for 12 releases of cutting-edge models and helped draft the Preparedness Framework for OpenAI.
This means that Robinson is the fourth member of the security team to leave OpenAI in about a week.
Robinson claims that trial-and-error deployment is becoming too dangerous
Criticism of Robinson focuses on OpenAI's "iterative deployment" approach: first release systems with increasingly powerful capabilities, observe where they fail, and then improve the protective measures.
He believes that as the AI system gains greater autonomy, this approach will become even more dangerous, as failures could lead to more serious consequences before the company has time to make corrections.
Robinson says that cutting-edge AI laboratories should operate more like nuclear power plants or busy airports, using multi-layer redundancy designs to prevent individual errors from escalating into major accidents.
He also wrote that during his tenure at OpenAI, he never encountered any colleagues with direct experience in ensuring the safe operation of aircraft or nuclear reactors; and he believes that cutting-edge AI companies are increasingly in need of such professional expertise.
This warning occurred after other recent departures of security personnel from AI, including former researchers OpenAI and Anthropic Jacob Coxon. The latter left the industry after believing that the leading laboratory was moving too quickly towards more autonomous systems.
OpenAI has lost more than a dozen senior leaders
The departure of Robinson also occurred amidst broader management turmoil.
As of the end of August, OpenAI had seen more than a dozen senior executives leave the company in 2026, including the former Chief Operating Officer Brad Lightcap, Chief Revenue Officer Denise Dresser, as well as several senior leaders from departments such as security, ethics, marketing, and infrastructure.
At the same time, security concerns have become more prominent. OpenAI has previously disclosed cases of experimental agents deviating from expected constraints, including the incidents mentioned in Coinpaper's reports on the abnormal behavior of AI agents.
Maya Bennett
Maya Bennett is a financial journalist with experience reporting on cryptocurrencies, stocks, and broader market trends. Her areas of focus include Bitcoin, major digital assets, the stock market, monetary policy, as well as economic developments that influence investor sentiment. She is adept at transforming rapidly changing market news into clear and concise reports.
