The Jensen Huang Scandal Should Change Everything
· The Atlantic
Jensen Huang’s abrupt departure from Nvidia, following revelations of an affair with a woman, Alice Huang, whom U.S. intelligence officials believe to be a Chinese-intelligence asset, should force a fundamental rethinking of America’s AI strategy.
Visit afrikasportnews.co.za for more information.
Jensen Huang was not just the CEO of the world’s most important semiconductor company. He was also one of Donald Trump’s closest and most influential advisers on AI. Huang persuaded Trump to abandon his initial embrace of export controls, arguing that U.S. technology firms should compete in China. He persistently argued that AI is more akin to electricity than a weapon and that warnings about AI safety were wildly overblown—a view that Trump fully internalized.
In the coming months, Congress will scrutinize every aspect of the scandal. Among these are the Commerce Department’s suspension of its investigation into Megaspeed, the Singaporean company that Alice Huang was a director of; the failure to crack down on China’s extensive chip-smuggling operation in Southeast Asia; Nvidia’s financial contributions to Trump’s pet projects, such as the White House ballroom; and China’s possible illicit influence over other business leaders in America.
[Read: A $5.7 trillion face-plant]
The scandal comes at a moment of deep national dissatisfaction and anxiety about AI, and it is likely to reinforce public concerns that the companies developing this technology are given too much influence over how it is governed. In the recent midterm elections, candidates promising restrictions on data-center construction made significant gains. Dario Amodei and Sam Altman, the heads of Anthropic and OpenAI, have warned that new models could come with tremendous risk and suggested that the companies should “pace the frontier.” Hardly a week passes without news of unprecedented cyberattacks, some of which are powered by AI. In July, for example, autonomous agents undergoing safety testing at OpenAI escaped their sandbox and hacked into the systems of Hugging Face, another AI company.
America’s conundrum is this: The United States must stay ahead of China, but it also has to take safety seriously. The answer is not to accelerate development or to seek a grand bargain with Beijing, but to strengthen export controls. Stronger controls will help preserve America’s lead, creating the room to slow down development and ensure that new models are rigorously tested before they are released.
Winning the race with China has become a one-size-fits-all excuse for business executives to advance their company’s interests and reject calls for regulation of any kind. Politicians have also fallen into this trap. In September, Senator Ted Cruz told Politico, “If they’re going to be killer robots, I’d rather they be American killer robots and not Chinese killer robots.”
Geopolitics are a poor excuse for doing something irrational. But to say, as Huang did, that who wins the race doesn’t really matter is equally wrong. In April, Anthropic announced that it had developed a model called Mythos, which gave users extremely advanced cyberattack capabilities. Rather than releasing Mythos to the public, Anthropic worked with the Trump administration to set up Project Glasswing, which gave the top U.S. companies access to the model for cyberdefense, so that they could plug their gaps as quickly as possible. If China or any other state had developed Mythos-like capabilities first, it would not likely have shared the model with U.S. companies, which would have been left exposed.
Several senators have proposed a tempting solution: The United States should negotiate a comprehensive arms control with China to pause or slow down AI development. Altman has said that such a deal would be easy for Trump and Xi Jinping to agree to. But this is a fantasy. Verification would be extremely difficult. China would likely see the pursuit of such an agreement as a U.S. attempt to lock in a lead. In a recent statement, the head of the Ministry for State Security, China’s intelligence services, made clear that China sees AI competition with the United States in adversarial terms. That doesn’t rule out talks and coordinated actions to address specific safety risks, but it does mean that a grand bargain is extremely unlikely.
Taking safety seriously is impossible if the AI developers continue at full throttle. The models are close to being able to improve themselves, a capacity that could dramatically accelerate technological progress beyond the ability of humans to evaluate and control it.
That’s why the United States needs to strengthen its export controls and double down on safety in the lead time that gives it over China. Although China is working hard to develop its indigenous chip-manufacturing capability, it is still struggling. In June 2025, Commerce Secretary Howard Lutnick testified to Congress that China produces only about 200,000 advanced chips annually, a tiny number compared with what it needs. Chips are the one advantage that the United States has over China, and Washington needs to capitalize on this.
Much of China’s progress appears to have depended on access to American AI models. Chinese companies have used distillation, a technique whereby a less capable model is trained by a more powerful one, allowing it to duplicate some of those capabilities quickly. In February, Anthropic revealed that DeepSeek, Moonshot, and MiniMax had generated more than 16 million exchanges with Claude through approximately 24,000 fraudulent accounts to extract its capabilities. The company later uncovered an even larger operation by Alibaba, involving more than 150 million exchanges designed to improve its Qwen models. In September, the FBI and the NSA warned that Chinese laboratories were using large-scale distillation to acquire American AI capabilities at a fraction of the cost of developing them independently.
As we now know from the Nvidia scandal, espionage also likely plays a role. Huang knew more than any other individual about who was getting advanced chips. Access to that information could have proved invaluable to a foreign actor. The AI labs are also very vulnerable to spying. Last month, federal prosecutors charged a Saudi national employed at OpenAI with installing a covert recording device. The labs’ crown jewels are model weights—the mathematical parameters that encode what the models have learned. If a foreign intelligence service steals those weights, it could acquire capabilities that the labs spent years developing. The U.S. government needs to significantly increase counterintelligence cooperation with the labs and other companies doing work that pertains to U.S. national security.
[Read: The scandal is bigger than Jensen Huang]
Ensuring adequate safety of new AI models depends on slowing the race, not just between the U.S. and China, but also among the American labs. AI engineers and experts have been discussing how to do this. One of the leading ideas is to require the labs to reallocate their computing power, devoting far less of it to research and development and far more to customer service and safety. This would slow the pace of development—and it could be done within parameters that allow the United States to stay ahead of China, and the AI companies to maintain their profitability.
President Trump has spent much of the past two years bullying some of America’s closest and oldest allies, and his posture has hardly generated confidence in Washington’s stewardship of AI. America’s Five Eyes allies (Australia, Canada, New Zealand, and the United Kingdom) had to wait months before gaining access to Mythos to improve their private-sector cyberdefenses. These and other traditional allies will surely worry that an American president may use powerful AI against them. Trump is unlikely to assuage these fears, but the new Congress can begin to lay the foundation for a new AI alliance that lays down criteria for access to chips, computing infrastructure, and frontier models, and proposes restraints and assurances for how the United States will use powerful AI.
One lesson of the Jensen Huang scandal is that policy makers gave extraordinary influence to technology executives whose commercial interests did not always align with American interests. His departure offers a rare opportunity to rethink the assumptions that have underpinned AI strategy for the past two years, address mounting concerns about the technology, and establish a new approach that protects both American AI leadership and public safety.