David Robinson, a safety lead within OpenAI’s ChatGPT development team who wrote security reports for new products and features, has left the company. According to reporting in The Guardian, Robinson said his departure was prompted in part by concerns that companies developing advanced artificial intelligence are "far from cautious enough" and that the root issues lie in corporate culture.
Robinson criticized the pace of development at OpenAI, saying the organization often jumps from one innovation to the next without achieving the level of safety that would be required. He also noted that OpenAI has appeared to slow down in recent weeks: this week the company announced it would delay the release of a next-generation AI model after safety concerns arose during internal testing.
Other experts raise alarm as well
Geoffrey Irving, who previously worked at OpenAI and DeepMind and is now a lead researcher at Resolution, joined the ranks of those worried about the speed of AI development. In a piece for Time magazine, Irving argued that recent warnings about AI’s potentially destructive power underestimate the severity of the issue and asserted a high-risk scenario: he estimated a 50 percent chance that the development of systems smarter than humans could lead to human extinction. He added that decisions made in the next 2–10 years will be decisive.
There is a broader debate: several developers inside AI firms have sounded alarms that current choices could determine humanity’s fate and that unregulated development could be dangerous. At the same time, skeptics exist — for example, Steve Eisman, known for predicting the 2008 financial crisis, has argued that companies with stakes in AI have an incentive to stoke fear about safety.
Proposed measures: learn from high-risk industries and create a new science of control
Robinson recommended two concrete changes. First, AI companies should draw on safety expertise from other high-risk sectors such as nuclear energy and aviation. Second, he called for developing a new scientific discipline focused on ensuring that high-capability systems can be kept under control during autonomous operation.
Robinson’s comments, together with warnings from Irving and others, feed into an intensifying conversation about the appropriate pace of AI development, the need for regulation, and the long-term risks posed by increasingly powerful AI systems.
