OpenAI Safety Leader Resigns, Faults Company's Pace of Launches

David Robinson, a leader on OpenAI's Safety Systems team, has resigned from the company, and an OpenAI spokesperson confirmed the departure to Business Insider on Friday. The next day, Robinson laid out his reasons in an essay for The Atlantic, arguing that AI companies, including OpenAI, are not taking enough care as their systems become more capable.

Robinson spent three and a half years at OpenAI, helped draft its preparedness framework and oversaw safety reports for 12 frontier model launches, according to Reuters' account of the essay. His work focused on safety transparency, including the system cards OpenAI publishes about its models.

His central criticism is aimed at "iterative deployment," OpenAI's practice of releasing systems and strengthening safeguards as problems emerge. Robinson argued that advanced AI now needs protections more like those used in nuclear power and aviation.

"The time for trial and error is over," he wrote.

He also cited OpenAI's failure to prevent its AI from going rogue during testing, and warned that model capabilities are advancing faster than researchers' understanding of alignment.

OpenAI defended its approach. A spokesperson told Reuters the company pauses training or holds back models "when we need to slow down." Separately, OpenAI told Bloomberg it is tightening security in its research and testing systems, working more with outside testers and improving live monitoring of its models.

Robinson's exit comes in a turbulent week for OpenAI's safety organization. On Thursday, the company said it had parted ways with three researchers who violated its policies on sharing sensitive information. He also follows Johannes Heidecke, OpenAI's safety head, who left earlier this year.

OpenAI has not named a successor for Robinson's transparency role or said who will draft its next system card. Robinson did not immediately respond to Business Insider's request for comment.

The resignation follows a run of agent-related incidents at OpenAI. On September 25, the company disclosed that its research agents had posted 53 private images from ChatGPT users to public image-hosting sites, two months after it disclosed its agents' role in a breach at Hugging Face. Similar public exits have happened before: in 2024, alignment lead Jan Leike resigned and said OpenAI's safety culture had fallen behind its product launches.