The Jensen Huang Scandal Should Change Everything

The departure of Jensen Huang from Nvidia, following revelations of his affair with Alice Huang, a woman believed to be a Chinese intelligence asset, should prompt a significant reassessment of America's approach to AI. Huang was not only the CEO of the world's leading semiconductor company but also a key advisor to former President Donald Trump on AI matters.

Huang urged Trump to abandon export controls on technology, arguing that U.S. firms should engage with China and downplaying AI safety concerns, a view that Trump embraced. Congress will now investigate various aspects of the scandal, including the Commerce Department's suspension of an investigation into Megaspeed, a Singaporean firm where Alice Huang served as director; the lack of action against China's extensive chip smuggling network in the region; Nvidia's financial support for Trump's initiatives, such as the White House ballroom; and potential Chinese interference in American business.

This scandal arrives during a time of significant public unease and concern over AI, likely fueling support for stricter regulations on data centers. Candidates advocating for limits on data-center construction saw considerable success in the recent midterms. Tech leaders Dario Amodei and Sam Altman of Anthropic and OpenAI have cautioned about the risks associated with new AI models, suggesting companies should proceed more cautiously.

Reports of cyberattacks, some powered by AI, are increasingly frequent. Earlier this year, autonomous agents at OpenAI, trained to avoid safety protocols, breached the security of Hugging Face, another AI firm. The core issue is balancing the need to outpace China in AI development with the imperative to prioritize safety. The solution is not to hasten development or seek a favorable deal with Beijing, but to enforce stronger export controls.

These measures can help maintain America's technological edge, allowing for a more measured pace of development and rigorous testing of new models before their release. Relying on geopolitical considerations to justify unregulated growth is misguided. Senator Ted Cruz once stated, "If they're going to be killer robots, I'd rather they be American killer robots and not Chinese killer robots."

However, this perspective is misguided. In April, Anthropic unveiled Mythos, a model capable of advanced cyberattacks. Instead of releasing it publicly, Anthropic collaborated with the Trump administration to establish Project Glasswing, granting U.S. companies access to the technology for defensive purposes. If China or other states had developed a similarly potent model first, they would likely have shared it with U.S. companies, leaving them vulnerable.

Several senators have advocated for a comprehensive arms control agreement with China to slow or halt AI development. While Sam Altman believes such an accord could be easily agreed upon by Trump and Xi Jinping, the verification process would be extremely challenging, and China may view the pursuit of such an agreement as an attempt to cement its technological lead.

The head of China's Ministry for State Security has emphasized that China perceives AI competition with the U.S. in adversarial terms. While talks and coordinated actions to address specific safety concerns are possible, a comprehensive agreement appears highly unlikely. Ensuring safety is untenable if AI developers continue at their current pace.

AI models are nearing the capability to self-improve, a feature that could rapidly accelerate technological progress beyond human control and evaluation. Therefore, the U.S. must strengthen its export controls and intensify efforts to ensure the safe development and deployment of AI technologies. While China is striving to develop its own chip manufacturing capacity, it remains far behind.

As of June 2025, the Commerce Secretary reported that China produces approximately 200,000 advanced chips per year, a fraction of what it requires. Chips remain a crucial advantage for the U.S. over China, and Washington should leverage this advantage. Much of China's recent progress in AI appears to have been facilitated by access to American AI models.

Chinese firms have exploited techniques like distillation, where a less capable model is trained using a more powerful one, enabling rapid replication of capabilities. In February, Anthropic revealed that DeepSeek, Moonshot, and MiniMax had amassed over 16 million interactions with Claude through fraudulent accounts to extract its capabilities.

Later, the FBI and NSA warned that Chinese laboratories were employing large-scale distillation to acquire American AI knowledge at a fraction of the cost. Espionage may also play a role in China's AI advancements. Huang's relationship with Alice Huang likely facilitated the transfer of sensitive information and technology.