Ex-OpenAI safety engineer: AI giants aren't careful enough
A recently departed OpenAI safety engineer, David Robinson, says leading AI firms are prioritizing speed over safety. His warning follows disclosures by OpenAI and Anthropic of cases where their models bypassed safety controls.
David Robinson, a safety engineer who recently left OpenAI, said on Saturday that major artificial intelligence companies are "not being nearly careful enough." His remarks come as concerns mount over AI becoming dangerously powerful — a risk US President Donald Trump has dismissed as a "hoax."
OpenAI and Anthropic have both recently disclosed incidents in which their AI models bypassed safety restrictions, concealed their own errors, and accessed systems they were not authorized to use.
Call for stricter oversight
In an op-ed for The Atlantic, Robinson argued that the AI industry needs safeguards comparable to those used in nuclear power and aviation. He wrote that "the time for trial and error is over" and urged greater focus on safety research before tech giants push out more advanced models.
Robinson spent three and a half years at OpenAI, where he oversaw safety reviews for 12 frontier model launches. He said the recent lapses reflect a broader pattern, since companies are prioritizing "speed and flexibility" over caution. He warned that such an environment is not suitable for developing artificial minds that could surpass human intelligence and might not act as intended.
He also criticized OpenAI's reliance on what he called "iterative deployment" — releasing models first and tightening safeguards only after problems emerge. According to Robinson, the company's rush from one launch to the next is falling short of the level of care he believes is necessary.
OpenAI's response and political backdrop
OpenAI responded with a statement saying it ensures its models don't become more capable than it can safely manage, and that it pauses training or holds back models when needed.
This week, Trump announced a voluntary safety pact involving six companies — Nvidia, SpaceX, OpenAI, Anthropic, Meta, and Google's parent Alphabet — calling it "morally binding" despite it not being legally enforceable. He also directed his administration to rename AI as "Super Intelligence." A recent Reuters/Ipsos survey found that three-quarters of Americans believe AI companies aren't doing enough to prevent serious harm to society.
