On October 3, local time, according to US sources, David Robinson, head of the OpenAI security system team, has resigned. An OpenAI spokesperson said that Robinson left the company last week. Previously, he was also responsible for the company's policy planning and participated in artificial intelligence security transparency work, including the development and release of model “system cards.” Robinson published an article in “Atlantic Monthly” saying that he believes that the current development method of artificial intelligence companies is “unacceptable”. The problem is not only specific safety rules and laws, but also corporate culture. He said that OpenAI has been using an “iterative deployment” method for a long time by finding problems and then continuously improving security protection measures, but as the capabilities of artificial intelligence systems continue to increase, this practice may have an increasing impact on security errors. Robinson also mentioned that OpenAI has previously experienced security incidents such as unexpected operation of artificial intelligence agents and models that bypass internet access restrictions, and believes that related issues are related to the industry's rapid development and frequent trial and error working methods. Earlier, OpenAI said it had fired three security researchers for “handling sensitive materials rather than reporting security issues.”

On October 3, local time, according to US sources, David Robinson, head of the OpenAI security system team, has resigned. An OpenAI spokesperson said that Robinson left the company last week. Previously, he was also responsible for the company's policy planning and participated in artificial intelligence security transparency work, including the development and release of model “system cards.” Robinson published an article in “Atlantic Monthly” saying that he believes that the current development method of artificial intelligence companies is “unacceptable”. The problem is not only specific safety rules and laws, but also corporate culture. He said that OpenAI has been using an “iterative deployment” method for a long time by finding problems and then continuously improving security protection measures, but as the capabilities of artificial intelligence systems continue to increase, this practice may have an increasing impact on security errors. Robinson also mentioned that OpenAI has previously experienced security incidents such as unexpected operation of artificial intelligence agents and models that bypass internet access restrictions, and believes that related issues are related to the industry's rapid development and frequent trial and error working methods. Earlier, OpenAI said it had fired three security researchers for “handling sensitive materials rather than reporting security issues.”