David Robinson Leaves OpenAI, Citing Lack of Consistency

David Robinson, former head of security systems at OpenAI, says he left the company due to a disconnect between its cautious rhetoric and the actual pace of development. OpenAI emphasizes its security practices and calls for regulating the "frontier" of AI.

OpenAI Emphasizes Caution and Safety Mechanisms

OpenAI claims to be strengthening its security practices as its models become more capable and states that it is working to anticipate risks associated with future advancements, according to a spokesperson. The company specifies that it is careful not to exceed its capacity to manage and secure its systems, and that it can suspend or hold back models if necessary to slow their development. In September, Jakub Pachocki, OpenAI's chief scientist, highlighted the need for extreme caution regarding recursive self-improvement, a situation where an AI could improve itself. OpenAI did not immediately respond to a request for comment.

A Departure Motivated by a Discrepancy Between Rhetoric and Practice

David Robinson, who led OpenAI's security systems team, explains that he felt a cognitive dissonance that prompted him to leave the company last week. He believes that OpenAI was not cautious enough and that the company's public discourse did not align with its internal actions. Robinson reports that despite his repeated warnings about the capabilities of new models and his desire to slow down development, OpenAI continued to maintain a rapid internal pace, continuing to train and deploy models he deemed dangerous. He describes an intense atmosphere, with teams working at a breakneck speed, and compares the internal coordination to a kind of ballet driven by adrenaline. He detailed these observations in his first interview after his departure.

Calls for Slowing Down: From Employee Letters to Altman's Remarks

In July, several hundred OpenAI employees signed an open letter titled "Pacing the Frontier," urging the government to impose regulations to slow down AI development. In September, Sam Altman, CEO of OpenAI, stated that he shared the position of Dario Amodei, CEO of Anthropic, who advocated for coordination among companies to slow down AI development. Altman also wrote on X that the frontier of AI needs to be regulated, specifying that this topic has been at the center of internal discussions at OpenAI in recent weeks.