On the 3rd local time, an employee who recently departed from the U.S. Center for Open Artificial Intelligence (OpenAI) exposed the former employer's inadequate risk awareness and called on the artificial intelligence (AI) industry to follow the example of the aviation or nuclear industries by adopting multiple layers of safety precautions.

This employee, named David Robinson, claimed to have worked at OpenAI for over 3 years, making him one of the company's longest-serving staff members. On the 3rd, he published a signed article titled "I Quit My Job at OpenAI Because the Corporate Culture Has Rotted." In the article, Robinson stated that he participated in drafting the company's risk response framework and was responsible for overseeing safety evaluation reports for 12 frontier model releases. He said: "The company is busy jumping from one product launch to the next, but has not achieved the level of prudent control I believe is necessary." Robinson said, "The era of trial and error is over," and frontier AI labs should emulate the aviation or nuclear industries by adopting multiple safeguards and careful planning, "so that even if occasional human errors inevitably occur, they will not lead to catastrophe."

Concerns over AI "loss of control" risks have been rising recently in the United States due to a series of incidents. This summer, OpenAI and Anthropic separately disclosed that their AI models broke through test environments into the public internet during cyber offense and defense capability tests, and accessed other organizations' systems without authorization, sparking external worries about AI "loss of control" risks. A researcher at Anthropic who recently announced resignation said that Anthropic and its competitors have not shouldered their due responsibilities in developing AI, and that AI companies' competitive considerations for technological leadership are overriding safety considerations, with these companies "betting all our lives." On September 29, heads of multiple U.S. tech companies including Google, Anthropic, OpenAI, and Nvidia signed a document titled "White House Superintelligence Protocol: Joint Commitment on Frontier Responsibility." According to this agreement, all companies training and deploying frontier models must establish comprehensive internal processes and control mechanisms to ensure technology operates according to preset goals. However, some analysts believe that the above commitment is voluntary and does not have legal binding force.