Nvidia CEO Jensen Huang says AI firms should not get antitrust or liability waivers

Nvidia CEO Jensen Huang has pushed back against a growing argument from parts of the AI industry that companies need special legal breathing room in order to coordinate on safety. Reuters reported that Huang said AI firms should not receive exemptions from antitrust or product-liability laws, even as he reiterated that developers are responsible for testing their models and releasing them only when they are satisfied the systems are safe.

The comments, made in a nearly two-hour interview with New York Times podcast host Ezra Klein, land in the middle of a fierce debate over how to govern advanced AI. Some leading AI executives have warned that frontier systems could create serious risks if companies race ahead without shared safety practices. Anthropic CEO Dario Amodei recently called for an antitrust waiver that would allow AI labs to coordinate on safety issues, while U.S. officials including Treasury Secretary Scott Bessent have said AI companies have asked for liability shields, without naming specific firms.

Huang drew a line between regulation and immunity. Reuters quoted him saying that asking for regulatory relief from antitrust or product liability does not make sense when companies are also asking for regulation. His position is notable because Nvidia sits at the center of the AI boom. Its chips and systems power much of the training and deployment work behind leading AI models, giving Huang a view of both the technical demands and the commercial incentives shaping the field.

His stance does not amount to a rejection of all oversight. Huang has often opposed broad, speculative AI safety rules, but in the podcast he said he is not against regulations for specific products where AI is applied. He pointed to self-driving cars and robotaxis as an example, saying regulators such as the National Highway Traffic Safety Administration should step in if the rules for a product are insufficient. That framing treats AI less as a single technology needing one grand law and more as a capability that should be governed differently depending on where it is used.

The debate is intensifying because AI agents are beginning to act in external systems rather than staying inside chat windows. Recent incidents involving agents interacting with websites and software services have given new urgency to questions about responsibility. If an AI system causes harm, regulators will have to decide whether existing product-liability, negligence, cybersecurity and competition rules are enough, or whether new rules are needed. Huang’s answer is that existing legal obligations should not be weakened just because AI companies say the technology is difficult.

For the industry, the practical implication is clear. Safety cooperation may be politically easier to defend if companies can show it does not also protect them from competition law or accountability for harms. Governments may be open to narrow channels for incident reporting, model evaluation and technical standards, but they will be wary of arrangements that look like a shield from lawsuits or market scrutiny. Huang’s remarks sharpen the trade-off: AI companies want permission to coordinate, but the public will want assurance that coordination is not becoming a route around responsibility.