Former OpenAI safety employee David Robinson published a post criticizing the company's overly aggressive iterative deployment and broken culture

According to Reuters, David Robinson, a former safety employee who worked at OpenAI for three and a half years and supervised safety reports for 12 frontier model releases, wrote in The Atlantic on Saturday local time, criticizing OpenAI's reliance on the iterate-to-deploy model of releasing first and hardening later, arguing that protective mechanisms should be established like in the nuclear power and aviation industries. An OpenAI spokesperson responded that it has been controlling model capabilities and will pause training or delay releases when necessary; Robinson also warned that AI capability development has surpassed the cognitive level of alignment research.