David Robinson, a former OpenAI safety employee who resigned after three and a half years, published an article in The Atlantic criticizing the company's development culture. He argues that the firm's focus on rapid deployment increases system failure risks and lacks sufficient caution.
Robinson, who previously drafted risk frameworks and oversaw safety reports for 12 model releases, suggests that AI safety protocols should mirror standards used in the nuclear and aviation industries. He claims the current iterative deployment model, which releases systems before fully securing them, is inadequate.
The former employee warns that AI capabilities are advancing faster than researchers' understanding of alignment, which ensures systems act according to human values. This critique adds to ongoing industry debates regarding the speed of development at companies like OpenAI and Anthropic.
In response, an OpenAI spokesperson stated that the company monitors its models to ensure capabilities remain within safe control limits. The representative noted that OpenAI pauses training or delays releases whenever it determines that slowing down is necessary.
