OpenAI safety veteran warns of systemic failure in AI development

Robinson argues that slapping superficial regulations onto existing development cycles is a futile gesture. He describes a Silicon Valley environment defined by extreme confidence, where a perpetual sprint toward larger models ignores potential failures. According to his analysis, the industry must shed its insular, move-fast-and-break-things mentality in favor of a profound, institution-wide humility.

To mitigate these dangers, Robinson advocates for a shift toward the rigorous safety standards seen in high-stakes sectors like aviation or nuclear energy. He suggests that frontier labs require layers of redundancy and slow, deliberate planning to ensure that inevitable human error does not escalate into a catastrophe. His departure follows a pattern of internal dissent, joining a growing exodus of researchers—including former Anthropic employee Jacob Coxon—who have publicly questioned the reckless trajectory of current AI labs.