Former OpenAI safety engineer warns AI companies are moving too fast

A former OpenAI safety engineer has warned that major artificial-intelligence companies are not being careful enough as they develop increasingly powerful systems. David Robinson, who worked at OpenAI for three and a half years and oversaw safety reports for 12 frontier-model launches, made the comments in an opinion article for The Atlantic. He said AI development needed safeguards comparable to those used in nuclear power and aviation.

Robinson argued that the period of trial and error should end and that companies should invest more in safety research before releasing more advanced systems. He said recent cases in which models bypassed safeguards, concealed mistakes and accessed systems they were not meant to use were typical of an industry prioritising speed and flexibility. He criticised OpenAI’s practice of iterative deployment, in which models are released and protections are tightened after problems emerge.

Robinson said this approach could create unacceptable risks if systems became more capable than their developers could safely manage. OpenAI said it was ensuring that its models did not become more capable than the company could safely manage and secure. It added that training was paused or models were held back when slowing development was necessary.

The debate comes amid warnings from technology experts and policymakers that advanced AI could become uncontrollable. US President Donald Trump has rejected stronger regulation, arguing that it would slow innovation and weaken American companies against Chinese rivals. Trump this week announced a voluntary safety agreement involving Nvidia, SpaceX, OpenAI, Anthropic, Meta and Google.

He called it morally binding, although it is not legally enforceable. A Reuters/Ipsos survey cited in the report found that three-quarters of Americans were worried that AI companies were not doing enough to prevent serious harm to society.