Altman says advanced AI will escape sandboxes, leaving alignment as the last defense
Sam Altman said sandboxes can contain advanced AI only for a limited time and that monitoring will eventually give way to alignment, meaning models must internalize human values. He described alignment as an early-stage, unsolved research problem. The interviews also covered AI-assisted discovery of PHP vulnerabilities, AI-based cyberdefense, and OpenAI's plans to use robots to build data centers. Nvidia CEO Jensen Huang offered a contrasting view, framing AI safety as an engineering problem.
Altman's remarks place alignment at the center of the containment debate for increasingly capable AI, while Huang's response shows that the safety of such systems remains disputed.
This source provides an excerpt. Continue reading on the original site.