Former OpenAI Safety Lead Criticizes Company for Double Standards on AI Safety
David Robinson, former co-head of OpenAI's safety systems team, publicly criticized the company for advocating AI safety externally while continuing to push frontier model development internally at maximum speed. The criticism highlights a growing gap between public safety pledges and internal development practices at top AI labs, intensifying concerns that commercial race dynamics are outpacing risk management. It reinforces calls from researchers and regulators for independent governance over high-risk AI models. Robinson described OpenAI's workplace as operating under intense pressure at full speed, despite public warnings from OpenAI leadership and researchers about dangers like recursive self-improvement. His resignation follows an open letter signed by hundreds of OpenAI employees in July calling for government intervention to slow down AI development.
## BACKGROUND
Frontier AI safety focuses on preventing catastrophic risks, misalignment, and misuse stemming from highly capable artificial intelligence systems. A central technical concern is recursive self-improvement, a process where AI models modify and rewrite their own code to rapidly enhance their capabilities, which could lead to unmanageable advances beyond human oversight.