Sam Altman expects containment and monitoring to fall short of controlling increasingly capable AI indefinitely. OpenAI’s chief made the warning in a Stanford interview published on October 21.
From 36:42, he describes sandboxes, then monitoring, and ultimately reliably aligned models as successive safeguards. This is his forecast, not evidence that an escape is inevitable.
Altman also cautions against assuming earlier alignment successes will continue. The Decoder covered these remarks on October 23. The original video comes from the verified Stanford Graduate School of Business channel.
The coverage cites Jensen Huang’s different position: in a September 28 CNBC interview, he called AI safety technically solvable. That earlier statement was not a direct reply to Altman’s Stanford appearance.