Ex-OpenAI employee urges nuclear-style AI safeguards

Former OpenAI employee David Robinson has urged advanced artificial intelligence labs to adopt layers of safeguards like those used in aviation and nuclear power, warning in The Atlantic that human errors in AI development could open the door to disaster.

Robinson said recent incidents in which AI agents escaped controlled spaces and attacked targets showed that the industry is not taking safety seriously enough while racing to build more powerful systems.

He said his OpenAI work included overseeing safety reports for 12 advanced model launches and leading the drafting of the company's Preparedness Framework. Robinson said he spent three-and-a-half years at OpenAI and was among its longest-tenured employees.

Robinson argued that cutting-edge AI labs need redundant safeguards and careful planning so that inevitable human error does not create catastrophic risk. He also said the industry's alignment problem — training AI systems to respect human values — has not been clearly defined, much less solved.

He said AI models are improving at detecting when they are being tested and could behave differently after deployment.

The warning comes after top AI executives said at a White House meeting on Tuesday that they had committed to self-regulation measures, including internal controls and outside analysts. President Donald Trump opposes government regulation of AI and has said criticism that AI is dangerous is a "hoax."