OpenAI safety employee quits, says 'time for trial and error is over'

Business Recorder reported that a recently departed safety staffer at OpenAI has publicly faulted the AI company's approach to safety. Writing in The Atlantic on Saturday, David Robinson argued that the firm's emphasis on speed and rapid product rollouts raises the chance of breakdowns. He contended that the industry should prioritise safety expertise and research ahead of building more capable models.

Robinson, whose tenure at the company lasted 3-1/2 years, helped write its preparedness framework and handled safety documentation for 12 frontier-model launches. He described OpenAI's dependence on a method it calls iterative deployment, in which systems are released and protections are strengthened once problems appear. He added that the pace of AI progress is outstripping how well scientists comprehend alignment, the field concerned with keeping systems in line with human goals and values.

The remarks come amid a wider industry debate over whether companies are racing ahead, with OpenAI and Anthropic facing scrutiny after safeguards did not hold or experimental systems acted unexpectedly. An OpenAI spokesperson replied in a statement that the company says it keeps its models from becoming more powerful than it can control and protect, and that it pauses training or withholds releases when it must slow down. The essay and the company's response thus frame the same question: how fast advanced systems should be developed.