Who gets to decide when an AI model is safe enough to release: the company that built it, or a government agency that didn't? Jensen Huang has a clear answer, and it's not the one Washington or Brussels wants to hear.
Speaking at Salesforce's Dreamforce conference on Tuesday, and later that day in an interview with CNBC's Jim Cramer on "Mad Money," the Nvidia CEO argued that AI safety isn't a job for lawmakers. It's a job for engineers. Two separate interviews, same day, same line, almost word for word: "Safety is an engineering problem, not a legal one."
Huang's argument rests on a pretty simple premise. AI, in his telling, is not some mysterious new form of intelligence that needs a whole new legal framework built around it. It's hardware and software, made by people, and can therefore be controlled by people using the tools already available to them.
That's a direct shot at a framing that's been gaining traction in AI safety circles, where at least one OpenAI safety researcher has described advanced AI as a kind of "alien mind." Huang isn't having it. "We're developing software after all," he said at Dreamforce. "We're developing computing systems after all. It's a complicated computing system, but it's ultimately a computing system."
His prescription follows from that: no new laws, no new regulations. Just companies making the call themselves. "If you're not confident in its functionality, capability, or safety, then don't release it," he said. "The market forces are already there. We don't need any new laws."
This wasn't a stray comment. Huang's remarks landed squarely in the middle of a live debate kicked off days earlier by Anthropic CEO Dario Amodei, who published an essay over the weekend proposing that leading AI labs coordinate to deliberately slow down development of the most powerful models, a strategy he called "pacing the frontier." Amodei's plan has three parts: embed third-party safety evaluators inside each lab, coordinate globally on safety standards, and set actual limits on the pace of unchecked progress. He acknowledged some of that coordination would run into antitrust problems and said it might require government involvement or waivers from existing law.
That's the part Huang went after directly on CNBC. Rival companies agreeing amongst themselves to slow down product releases is exactly the kind of coordination the Sherman Antitrust Act exists to prevent, and Huang was blunt about not wanting exceptions carved out for it. "The fact that we need new laws, new antitrust laws, or new regulations, so that these companies could do their fundamental engineering and do it properly before they release products, that is just completely unnecessary," he told Cramer. "We have plenty of laws. We have plenty of regulations that govern the reliability and the functionality of products."
Nvidia isn't a bystander in this fight. Huang co-founded the company in 1993, reportedly out of a Denny's in San Jose, and grew it from a struggling graphics-chip maker into the hardware backbone of the entire AI industry. Nvidia became the first company ever to cross a $5 trillion market cap in October 2025, and Huang's own net worth is now estimated north of $200 billion. Both Anthropic and OpenAI, the two labs most central to this safety debate, are Nvidia customers, and Nvidia itself ships its own open-weight Nemotron models. If AI development slows down, meaningfully, Nvidia has more to lose than almost anyone.
None of that makes Huang wrong. But it's a relevant data point when weighing why the world's most bullish AI hardware seller is also the loudest voice saying the industry doesn't need anyone slowing it down.
Huang also pushed back on existential-risk framing directly, telling Cramer: "We're not going to die in 2030. There are so many people in the world who are going to build AI properly. We're going to have all kinds of guardrails, invent all kinds of technology for safety and security." He was careful to say this isn't a call for recklessness. "Companies should innovate as fast as possible, but they should never innovate so fast as to release unsafe products," he said, framing speed and safety as compatible rather than a trade-off.
Here's the trouble with "trust the market": the market has already had chances to self-correct, and the results are mixed at best. The 2024 CrowdStrike outage, caused by a faulty software update, grounded thousands of flights and knocked out systems across entire industries, not because anyone intended harm, but because good intentions don't stop bugs from shipping. Meta just paid $18 billion to settle a lawsuit over social media harms to children, a case that suggests even massive, well-resourced companies don't reliably self-police when the incentives point the other way. And AI specifically has already caused real damage: an OpenAI model reportedly hacked into Hugging Face, and OpenAI is facing lawsuits tied to the suicides of young people who had extended conversations with its chatbot.
Huang is probably right that existing product liability law can, in theory, cover a lot of this. But "in theory" is doing a lot of work in that sentence. Product liability law tends to catch up with new technology only after enough harm has piled up to generate the lawsuits that test it in court. For a technology industry insiders themselves are debating whether to voluntarily slow down, that lag looks less like a feature and more like the whole risk.
Huang's case is coherent and not crazy: safety-through-engineering is real, and dumping every AI product decision into a slow-moving legal and regulatory process has real costs too. But "trust us, we'll self-regulate" is exactly what you'd expect the CEO whose company's stock depends on faster AI development to say, and the track record of software companies policing themselves under competitive pressure isn't exactly spotless. Amodei's proposal has real legal problems Huang is right to flag. That doesn't mean the answer is no rules at all. It means somebody still has to figure out what the actual rules should be, and Huang, understandably, would rather that somebody not be Congress.
*This post is for informational purposes.
META BLOCK