Jensen Huang Says AI Safety Should Be Engineered, Not Regulated

Nvidia chief executive Jensen Huang pushed back on the idea that artificial intelligence needs its own set of government rules, arguing that AI is not a new kind of "alien mind" but simply hardware and software that companies can make safe through engineering. His comments, reported by TechCrunch, put the head of the world's most valuable chipmaker at the center of the debate over how — or whether — to regulate the technology his company supplies.

Huang's position rests on a straightforward claim: because AI systems are built from code and silicon, the firms that build them can test, constrain and fix them the way they handle any other complex product. That view aligns Nvidia with much of the industry, which has lobbied against broad new AI statutes and favored voluntary standards and targeted rules instead. It also sets him against lawmakers in the United States and Europe who have spent the past two years drafting binding requirements for high-risk AI systems.

The stakes are considerable for Nvidia, which sells the GPUs that train and run most of the leading AI models. Any regulation that slowed deployment or imposed heavy compliance costs would ripple through its order book, which has surged over the past two years on demand from cloud providers and AI startups. Huang has consistently framed AI as a general-purpose computing shift, comparable to past platform transitions, rather than an existential risk requiring a new regulator.

His argument lands as regulators move on several fronts. The European Union's AI Act is being phased in, with obligations for general-purpose models taking effect in stages, while U.S. agencies and states have advanced their own proposals and executive orders have shifted with administrations. Industry executives have largely favored a patchwork of sector-specific oversight over a single new agency, warning that rigid rules could freeze development while the technology changes faster than legislation.

Safety researchers and some policymakers counter that self-regulation has a mixed record, and that companies face commercial incentives to ship quickly. They point to the difficulty of auditing models whose behavior is hard to predict, and to the concentration of AI development inside a handful of firms. Huang's response is that engineering discipline — red-teaming, evaluations and guardrails built into products — is the practical path, not a new layer of law.

What happens next may depend less on the argument than on timing. If regulators in Brussels or Washington finalize rules that bind model developers, Nvidia and its customers will have to comply regardless of Huang's preference. The open question is whether voluntary engineering standards can move fast enough to satisfy governments that have already decided the technology needs a rulebook.