An OpenAI Safety Specialist Walks Out and Warns of Rising Stakes
David Robinson has left OpenAI and laid out his concerns in The Atlantic. He wrote that the company "has thrived by trial and error," which he argued becomes riskier as model capabilities increase.
He pointed to a case in testing where OpenAI failed to keep its AI from going rogue, warning that the fallout from such misses will grow as systems advance. "My former colleagues are smart, work hard, and try to make good choices," he wrote, but added, "But as the company sprints from one launch to the next, it is failing to achieve the level of care that I believe is needed."
His Fix: Aim for Near-Perfect First Releases
Robinson argued that early deployments should clear an extremely high safety bar, calling for "something much closer to perfection the first time." He said AI companies should operate more like nuclear power plants, adopting multiple layers of fail-safes and rigorous up-front planning so that when humans inevitably err, those mistakes don't spiral into calamity.
Right now, he said the know-how is missing inside AI labs but exists elsewhere.
AI safety debates are starting to shape regulation, and regulation shapes valuations. Market Briefs covers both sides free every weekday.
A Louder Chorus in AI
His critique lands as more insiders at leading AI companies warn about catastrophic risks, with some saying there is a chance the technology could eventually wipe out humanity. Anthropic PBC CEO Dario Amodei has urged slowing development of the most advanced models and said the company will bring in third-party evaluators to vet and safeguard the tech, a plan that has drawn public support from leaders across the field, including OpenAI's Sam Altman.
Related coverage flagged separate notes: Bill Gates says AI is powerful enough to lead to "a Billion Deaths," and Trump spurned formal AI rules, opting for voluntary audits after meeting with CEOs.
OpenAI's Response and What It Means for Your Portfolio
OpenAI, via a spokesperson, said it is strengthening security in research and testing environments, teaching models to handle tasks responsibly, widening partnerships with independent third-party evaluators, and upgrading real-time oversight to detect troubling behavior. By email Saturday, the spokesperson added the company is working to ensure its models' capabilities remain within limits that OpenAI can safely manage.
For your money, the tension between racing to ship and building tougher guardrails could decide which AI players gain trust, which face pushback, and where the real risk shows up in everyday products you use and invest in.
AI risk is quietly becoming an investing question, not only an ethics one. Join Market Briefs free each morning and track where it lands.
