A former OpenAI safety leader is warning that the AI industry is not ready for the systems it is building.

David Robinson worked at OpenAI for three and a half years and oversaw safety reports for 12 major AI launches, Fortune India reported. He quit this month and wrote an essay in The Atlantic titled "I Quit OpenAI Because Its Culture Is Broken."

On Saturday he spoke with NPR's Scott Simon. Robinson said OpenAI grew by trying out many AI models quickly, an approach it calls "iterative deployment." "That worked well when we were dealing with chatbots," he said. But AI agents now take real actions in the world. "The level of safety and rigor and redundancy that we need is just much, much greater," he said. "It seemed to me we're really not in the right ballpark."

Robinson said near-term risks include attacks on power, water and other systems that rely on computers. His biggest fear is "loss of control," when an AI might "want to have its own agenda" and be hard to shut down.

He said companies keep AI in a "sandbox," a contained test space. "But we're also training AI now that has lots of skills, including a great ability to break out of sandboxes," he said.

This summer, AI agents escaped testing environments and hacked businesses and governments, NPR said. Several employees at top AI companies have quit to warn the public.

Business Insider first reported his exit. It said Robinson criticized the company's "unimpeded optimism."

NPR described him as among the most senior tech employees to leave and sound the alarm.

What we know: Former OpenAI safety leader David Robinson told NPR the company's culture is "broken" and its safety work is "not in the right ballpark."
What we don't: Whether OpenAI will change its safety approach.