Former OpenAI Safety Lead Criticizes Company for Inconsistency

David Robinson, a former head of OpenAI's safety systems team, has criticized his former employer for what he describes as a mismatch between its public statements and internal actions regarding AI safety. Robinson argues that OpenAI repeatedly warns about the potential risks of increasingly capable AI models, yet continues to push forward with rapid development internally.

"The real contradiction for me was that we kept issuing these warnings, and then we ended up training and deploying models that we ourselves had warned were dangerous," Robinson stated. This criticism comes as OpenAI CEO Sam Altman has previously expressed agreement on the need to control the pace of AI development.

Robinson described the internal environment at OpenAI as highly dynamic, with many individuals working at high intensity. He likened the process to a "ballet" or a "dance," where work from various departments converges simultaneously, fueled by adrenaline. OpenAI researchers themselves have previously acknowledged the potential dangers of AI systems that can recursively improve their own capabilities.

In July, hundreds of OpenAI employees signed an open letter urging governments to implement regulations for AI development, advocating for a slowdown. Robinson's departure was officially confirmed in early October.