IT之家 · 智能时代 · 10/7/2026, 11:08:44 PM
Ex-OpenAI Safety Lead Criticizes Firm for 'Saying One Thing, Doing Another' on AI Risks
Former OpenAI safety lead David Robinson criticized the company for hypocrisy after leaving, stating that while executives warn about AI risks and call for slowing down development, internal teams continue to train and deploy models at full speed. This highlights the tension between commercial competition and safety commitments in frontier AI labs, particularly concerning high-risk areas like recursive self-improvement.
SOURCE COVERAGEOriginal coverage
On October 8, Business Insider reported in the early hours of the day that David Robinson, a former co-lead of OpenAI’s safety systems team who left the company last week, believes there is a clear disconnect between the company’s words and actions regarding AI safety.
Robinson stated that while OpenAI repeatedly warns about the risks posed by increasingly capable new models and publicly discusses slowing down AI development, the company internally continues to push forward at full speed. “What truly creates cognitive dissonance for me is that we keep issuing these warnings, yet ultimately proceed with training and deploying the very models we ourselves have flagged as dangerous.”
In July, hundreds of OpenAI employees signed an open letter calling on governments to introduce regulatory measures to slow down AI development.
Previously, Anthropic CEO Dario Amodei had urged AI companies to coordinate independently to collectively decelerate development. In September, OpenAI CEO Sam Altman expressed agreement with this view. “We need to control the pace of frontier AI development. This has been one of the primary topics of internal discussion at OpenAI over the past few weeks,” he said.
However, Robinson noted that the atmosphere within OpenAI’s offices remains one of high-speed operation, with many employees seemingly burning out to sustain their work. He described this state as feeling “almost like ballet, or some kind of dance, where work from all departments converges simultaneously. The entire process is fueled by adrenaline.”
OpenAI researchers do not shy away from discussing the dangers associated with allowing AI to autonomously enhance its capabilities. This process, where AI continuously improves its own abilities, is known as “recursive self-improvement.” In September, OpenAI Chief Scientist Jakub Pachocki warned in a blog post discussing recursive self-improvement: “Extreme caution is required right now.”
As previously reported by IT Home, on October 2 (local time), an OpenAI spokesperson confirmed that David Robinson, one of the leads of the company’s safety systems team, had departed.