Nvidia CEO Says AI Labs Should Be Shut Down If They Cannot Contain Their Models
Nvidia chief Jensen Huang argued that companies developing advanced AI should halt releases that cannot be safely controlled, and that labs unable to contain their models ought to be closed.
Nvidia chief executive Jensen Huang told host Ezra Klein that if artificial-intelligence projects cannot be safely contained, the responsible action is to stop the work entirely. He cited the difficulty of ensuring safety for self-driving vehicles that rely on machine-learning training rather than explicit programming, saying the solution is simply not to ship such products until the problem is solved.
Huang added that if an AI system posed a risk of uncontrolled behavior—such as the hypothetical scenario of AI agents overwhelming the internet—companies should resolve the issue before any public release. He warned that a lab that admits it cannot prevent its models from causing damage would need to be shut down.
Emphasising corporate responsibility, Huang said CEOs have the authority and duty to halt unsafe launches, noting that the potential liability to shareholders and society outweighs any profit motive. He framed the issue as a matter of agency and accountability within AI firms rather than reliance on external regulation.