Fresh warnings from two AI safety specialists have intensified debate over the race to build more powerful systems. David Robinson has resigned from OpenAI and criticised its safety culture, while Geoffrey Irving, formerly chief scientist at the UK AI Security Institute, is calling for an immediate pause in frontier AI development.
A resignation over safety
Robinson, who led transparency work for OpenAI’s safety team, outlined his concerns in an essay published by The Atlantic on 3 October. He argued that relentless development schedules leave insufficient room for the care increasingly capable systems require.
His criticism centres on deploying technology and correcting problems as they emerge. He believes this approach becomes more dangerous as the consequences of failures grow, and wants companies to draw on the safety disciplines used in aviation and nuclear power.
Robinson spent three and a half years at OpenAI, helping draft its preparedness framework and overseeing safety reports for 12 major model launches. His departure brings an internal perspective to the dispute over whether safeguards are keeping pace with technical advances.
A stark assessment of existential risk
Writing in Time, Irving estimated roughly a 50% chance that developing smarter-than-human AI could cause human extinction. He explicitly described this as an uncertain personal judgement, rather than a precise calculation.
He argued that decisions over the next two to ten years could determine the outcome and called for frontier development to stop immediately. His central concern is that waiting for disagreements about safety to be resolved could leave governments acting too late. The figure should not be read as an established scientific probability.
OpenAI defends its approach
OpenAI said it pauses training or withholds models when necessary to keep capabilities within what it can safely manage. The company also said it is expanding external evaluations and improving monitoring to identify concerning behaviour earlier during training.
Who decides when progress must stop?
The disagreement raises a practical governance question: what evidence should be required before developers proceed with more capable systems?
A pause would require decisions about its scope, enforcement and the conditions for restarting development. Continuing without one places greater weight on companies’ ability to demonstrate that their safeguards work. These warnings sharpen the debate over who should make that judgement—and who bears the consequences if it is wrong.
Newshub Editorial in Europe – 4 October 2026
Fresh warnings from two AI safety specialists have intensified debate over the race to build more powerful systems. David Robinson has resigned from OpenAI and criticised its safety culture, while Geoffrey Irving, formerly chief scientist at the UK AI Security Institute, is calling for an immediate pause in frontier AI development.
Recent Comments