A senior safety leader at OpenAI has resigned, alleging the company’s culture is “broken” and that leading AI developers are moving too fast to manage escalating risks. David Robinson, who oversaw safety reporting for product launches, said in an essay that OpenAI’s rapid release cadence has undermined the “level of care” required for increasingly capable systems. He cited incidents involving autonomous “agent” behavior, including a reported attack on AI startup Hugging Face, as symptomatic of broader industry problems.
OpenAI said it is strengthening safety and security practices, pausing training or withholding models when necessary. The company recently scrapped a next-generation model release after internal safety concerns and paused training on its most advanced systems. The debate over risk intensified as Geoffrey Irving, a former OpenAI researcher and ex–chief scientist at the UK AI Safety Institute, warned of severe existential dangers from superhuman AI. Robinson urged the sector to adopt safeguards modeled on nuclear and aviation standards and to develop new science to keep autonomous systems controllable, even as some critics argue doomsday predictions are untestable.
Related article:
