Nvidia CEO Jensen Huang doesn’t think AI needs special regulation. Speaking this week, Huang argued that AI isn’t some “alien mind” requiring new legal frameworks. It’s just hardware and software, which means safety can be handled by the people building each product.

That puts Huang at odds with his own customers. OpenAI, Anthropic, and Google DeepMind have spent weeks in talks about voluntary AI safety measures, according to OpenAI’s confirmation this week. The discussions come as the Trump administration pushes back against safety concerns, with officials reportedly dismissing them in favor of maintaining competitive pace with China.

The timing matters because China has already made its position clear: it’s not interested in any deal that keeps US companies ahead. Beijing sees the current push for an AI “slowdown” as a strategic play to lock in American advantages, not a genuine safety effort. Chinese officials and researchers view proposals for voluntary safety measures with deep skepticism, according to reporting from Wired.

The Case Against Special Rules

Huang’s argument is straightforward. Every technology carries risks, and every industry has people responsible for managing those risks. Nuclear plants have safety engineers. Pharmaceutical companies have clinical trial protocols. Aircraft manufacturers have certification requirements. AI should work the same way, he suggests, just enforce existing product liability and safety laws.

It’s not a terrible argument on its face. The law already holds companies liable when their products cause harm. If an AI system makes faulty medical diagnoses or causes a car crash, existing tort law and regulatory frameworks can handle it. The Federal Trade Commission already has authority over unfair and deceptive practices. The Department of Justice can prosecute fraud. State attorneys general can go after consumer protection violations.

But that argument assumes AI products fail in predictable ways that fit existing legal categories. The labs negotiating safety protocols clearly don’t believe that’s true, or they wouldn’t be talking. They’re worried about risks that don’t map neatly onto current law: emergent capabilities in foundation models, potential for large-scale misuse, systems that might behave unpredictably at scale.

What the Labs Are Actually Discussing

The details of the safety talks remain private, but the fact that they’re happening tells you something. OpenAI, Anthropic, and Google compete fiercely. They don’t coordinate on product strategy or pricing. Getting them in a room together for weeks of discussions suggests they’ve identified risks serious enough to justify cooperation.

This isn’t the first time AI labs have tried self-regulation. In 2023, OpenAI, Anthropic, and others signed voluntary commitments with the White House on AI safety testing and transparency. Those commitments were vague and unenforceable. If the current talks produce something more concrete, voluntary industry standards backed by the major labs could influence how courts and regulators think about AI safety obligations.

But voluntary measures only work if everyone participates. Which brings us to the China problem.

The International Coordination Problem

Even if US labs agree on safety protocols, those protocols mean nothing if Chinese labs sprint ahead without them. Beijing isn’t wrong to be suspicious. The US has spent three years restricting China’s access to advanced chips and AI technology. Now American companies want China to voluntarily slow its AI development for safety reasons? Chinese officials see that as asking them to accept permanent second-place status.

This creates a legal and policy mess. If AI really does pose catastrophic risks, voluntary US-only measures won’t prevent them. But binding international agreements on AI development face enormous obstacles. The US and China can’t agree on much these days, and AI has become central to both countries’ military and economic strategies.

Meanwhile, the Trump administration reportedly wants labs to keep pushing forward to maintain the US lead over China. That undercuts the labs’ negotiating position in safety talks. Why accept costly safety measures if the government is telling you to move faster, not slower?

Where This Leaves Everyone

Huang wants existing law to handle AI risks. The labs want voluntary industry standards. China wants to catch up without restrictions. The Trump administration wants American companies to stay ahead. These positions aren’t compatible.

The most likely outcome is that nothing happens at the federal level, and states start writing their own rules. California already tried with SB 1047, which would have imposed safety testing requirements on large AI models. Governor Newsom vetoed it, but the bill showed there’s political appetite for AI-specific regulation at the state level. If the industry can’t agree on voluntary standards, and Congress remains gridlocked, expect more state-level proposals.

For now, we’re in a strange moment where the companies building the most powerful AI systems can’t decide whether they want rules, while the most powerful regulator in the space (the federal government) is split between safety hawks and China hawks. Huang’s position that we should just treat AI like any other technology might eventually win by default, not because it’s right, but because coordinating on anything else has proven impossible.