Nvidia CEO: AI Labs Must Shut Down If They Can't Contain Models

Nvidia CEO Jensen Huang has staked out a position on AI safety that rejects apocalyptic scenarios while demanding accountability: if artificial intelligence companies genuinely believe their technology could escape their control and harm humanity, they should shut their labs down rather than release it.

In a recent interview on the Ezra Klein Show, first reported by Gizmodo, Huang dismissed concerns about AI posing an existential threat to humanity. But he also outlined a straightforward framework for companies that claim otherwise.

The containment principle

Huang drew an analogy to autonomous vehicles to illustrate his point. If engineers building a robotaxi encounter a difficult condition they cannot solve—because the system is trained rather than programmed—the answer is simple: don't ship it.

"If [AI labs] say the alternative, which is: There is no way to contain our experiments, there's just no way; when we test our A.I. models, it will get out, and it will damage the world, then I think the answer is that we have to shut the labs down," Huang said.

This comes as some AI leaders have warned of near-term risks. Anthropic CEO Dario Amodei recently suggested that AI agent swarms could threaten internet infrastructure within months without a development slowdown. Huang's response places the burden squarely on company leadership.

Why it matters

Huang's comments highlight a growing tension in the AI industry between companies that claim their technology may be advancing beyond their ability to control it and the expectation that they maintain responsibility for what they build. His framework challenges the narrative that AI development is somehow inevitable or beyond human agency, reasserting that CEOs have both the power and obligation to halt releases they deem unsafe.

Agency and accountability

The Nvidia chief emphasized that AI laboratories and their executives possess full agency over their decisions. "These are companies with agency. These are CEOs with agency," he said. "If I believe that I'm about to launch a product that is unsafe, it is completely in my ability, my power and my responsibility, and I'm incentivized to do so, to not launch the product."

However, Huang's reasoning for why labs should exercise restraint reveals a market-focused perspective. He cited shareholder concerns, civil and criminal liabilities, and potential damage costs as primary motivations for caution—framing the decision in terms of business risk rather than purely ethical obligation.

"Because the cost to humanity, the damage is too great. The shareholder, the liabilities—it could be civil liabilities, it could be criminal liabilities. I mean, the liability's incredible," Huang explained.

The approach essentially amounts to a "don't build the torment nexus" philosophy: if you know something will cause harm and cannot prevent it, simply don't create it. While the logic is sound, grounding it primarily in liability concerns rather than human welfare underscores how corporate incentive structures shape AI safety discussions.

The details were first reported by Gizmodo.