David Robinson, a former leader of the safety systems team at OpenAI, resigned from the company last week. In a recent statement, he criticized the organization for a discrepancy between its public warnings about AI risks and its internal development practices.
Robinson noted that while OpenAI publicly discusses slowing down AI development to mitigate potential dangers, the company continues to train and deploy advanced models at full speed. He described the internal environment as a high-pressure, adrenaline-fueled race to meet development goals.
The issue of AI safety has been a subject of internal debate at OpenAI. In July, hundreds of employees signed an open letter calling for government regulation to manage the pace of AI progress. CEO Sam Altman also acknowledged in September that the company needs to control the development rhythm of frontier AI.
Concerns regarding AI safety extend to the concept of recursive self-improvement, where AI systems enhance their own capabilities. In September, OpenAI Chief Scientist Jakub Pachocki warned that extreme caution is necessary when dealing with such processes, despite the company's aggressive development schedule.
