Nvidia CEO Jensen Huang has a pointed message for the engineers building today’s AI systems: keeping those systems safe falls on them first. Speaking at Salesforce’s Dreamforce conference, Huang said new laws are not needed to govern artificial intelligence, arguing that AI safety should be handled through engineering and existing rules. “Safety is an engineering problem, not a legal one,” Huang said.The Nvidia chief also brushed aside the idea that AI is some new form of “alien mind,” as at least one OpenAI safety researcher has described it. To him, it is just hardware and software built by humans, which means it can be controlled by humans and existing laws. He has been building the hardware behind AI since well before ChatGPT existed, and he sees the technology in plain terms. “It’s a complicated computing system, but it’s ultimately a computing system,” he said.
Jensen Huang says AI needs no new laws or regulations
Huang’s argument leans on the market. He said market pressure should be enough to stop companies from releasing unsafe AI products, so there is little reason for governments to step in. “We don’t need any new laws. We don’t need new regulations,” Huang said.He also rejected the popular framing that companies must pick between speed and safety. Huang called it a false choice and said firms could have both at the same time. His advice was to push ahead at full pace, but if a company ever feels out of control or unsure a product is safe, to “take a pause and make sure you get it right.” He said companies should not release products if they are not confident in their functionality, capability or safety.Huang made no secret of his appetite for growth either. “I’m more ambitious than ever,” he said, adding that there is no ceiling for Nvidia, for every industry and for every country.
Mark Zuckerberg makes a similar case for self-policing AI labs
Huang is not the only tech chief putting the burden on companies. Meta CEO Mark Zuckerberg laid out a comparable view in a recent post, arguing that every AI lab has both the responsibility and the incentive to train its models safely.His reasoning is commercial. People will not use AI agents that ignore their instructions or work against them, so alignment is fast becoming a feature that separates one model from another. Labs that neglect it, he wrote, will fall behind. He also flagged legal liability as a strong reason for companies to avoid shipping harmful models.Zuckerberg pointed to Meta’s own record. The company delayed its Muse model by several months to focus on safety and security, without first asking rivals to do the same. He added that Meta Superintelligence Labs already brings in independent evaluators, and that Meta has committed most of its compute to serving people rather than racing towards recursive self-improvement. Other labs, he said, can make the same call.
Critics say AI safety cannot wait for courts to catch up
Not everyone is convinced. Existing product liability laws could cover AI, but testing that theory depends on enough cases making their way through the courts. That means limits may get drawn only after something breaks. Recent examples include the 2024 CrowdStrike outage that disrupted thousands of flights, along with an OpenAI model’s breach of Hugging Face and lawsuits against OpenAI linked to user deaths following extended chatbot conversations.Huang also did not address industry self-regulation, a path he has instead pursued by backing open-weight models as a competitive counterweight to proprietary AI labs. Meanwhile, Microsoft CEO Satya Nadella, speaking at the All-In Summit on Monday, said China should share the same AI safety concerns as the US, since it faces the same hacking risks.That puts the weight of Huang’s argument squarely on engineering teams. It holds up only if companies actually hit pause when their own engineers raise a flag, even when growth targets point the other way.
