As China’s open-weight AI models become more capable, the Trump administration is weighing how far the US should go in restricting them. The debate is not concentrated in one office. It is spread across several departments and agencies, with officials taking different views on national security, trade, innovation, and the best way to keep the US ahead.
One senior White House official described the situation to WIRED this way: “It’s not an argument with two sides, it’s an argument with ten sides.” That captures the central tension. The administration is not simply choosing between regulation and no regulation. It is deciding which risks matter most, which tools to use, and which voices will shape the final policy.
A Small Circle With Competing Instincts
The source of influence is fragmented. With limited interagency coordination, policy is expected to be shaped heavily by the people who have the most sway with President Donald Trump. That puts a small set of officials at the center of the US approach to Chinese AI.
Commerce Secretary Howard Lutnick has become an important figure because the Commerce Department oversees export controls through the Bureau of Industry and Security. According to people familiar with the matter, Lutnick has considered incentives for top US labs to create open-weight models of their own as a counterweight to China. He has also spoken with leaders at several AI labs.
Lutnick appears to occupy a middle position. He imposed export controls on Anthropic, but he has also been less restrictive than others in the White House. On Thursday, he played down Moonshot AI’s Kimi K3 model in an X post, saying his team found it underperformed leading US frontier models on key benchmark tests.
Working under Lutnick is Arvind Raman, acting director of the Center for AI Standards and Innovation, or CAISI. The center sits inside the Commerce Department and functions as the AI industry’s main contact point with the government. Raman took the role last week after Chris Fall abruptly resigned.
CAISI, Anthropic, and the Safeguards Debate
CAISI has already been involved in a major AI safety dispute. Before leaving, Chris Fall had spent weeks trying to get Anthropic to strengthen safeguards against jailbreaks of its most powerful Fable 5 model. WIRED previously reported that CAISI staff met with Anthropic’s technical teams for weeks while both sides worked through new protections considered strong enough for Fable 5 to return online.
That episode shows one side of the administration’s AI agenda: controlling the risks of powerful US models. But the debate over Chinese open-weight AI adds another layer. Officials are not only considering what American companies should be allowed to release. They are also asking how Chinese labs might use US technology, and whether Washington should respond with trade tools, sanctions, or other restrictions.
Hardline Voices Focus on China and Distillation
National Cyber Director Sean Cairncross has taken a tougher position toward Chinese AI labs. He has a major role in potential regulation of Chinese AI and is empowered at the White House to develop policy addressing possible national security risks. He helped assemble President Donald Trump’s June 2 executive order, which created a framework for assessing the most powerful AI models.
Cairncross previously worked as a political campaign lawyer and was most recently at the Republican National Committee. He does not have tech or AI experience. Even so, current and former staff credit him with engaging seriously with national security questions and giving staff room to work without micromanaging each issue like a software engineer might.
One concern for Cairncross and his chief of staff, Lara Smith, is distillation. In this context, the concern is that Chinese labs could train their models using the outputs or capabilities of leading US models. White House spokesperson Elizabeth Huston said the administration’s discussions about Chinese AI are focused on strengthening US AI.
Treasury Secretary Scott Bessent has taken perhaps the most aggressive stance toward Chinese AI and distillation. As the senior Trump official handling US-China trade relations, Bessent has threatened Chinese labs with sanctions or inclusion on the Entity List, a US government trade-restriction list. In a post on X last week, he described distillation as “IP theft.”
Bessent has increased his role in AI policy ahead of Chinese president Xi Jinping’s visit to the White House in September and an APEC summit in November. According to a person familiar with his thinking, Bessent has told White House Chief of Staff Susie Wiles and others that the US, Europe, Japan, and other Western allies need to move faster on open weight models or risk losing the AI race to China.
The Outside Adviser and the Political Gatekeeper
David Sacks, the former AI czar, remains a powerful adviser even after leaving that role in March. The tech investor still has a direct line to President Donald Trump and has argued for a hands-off approach to AI. He also intervened at the last minute to reduce some regulatory provisions in the June 2 executive order.
Sacks has extended that laissez-faire view to Chinese open-weight AI models. He uses his X account, which has 1.6 million followers, to influence the administration from outside government. On July 23, he wrote that “The Kimi Panic needs to stop,” arguing that American frontier models remain ahead and that continued releases will help maintain the lead. The post had nearly 500,000 views.
Inside the White House, Susie Wiles plays a decisive filtering role. As Trump’s top political aide in the process, she reviews and scrutinizes proposals before they reach the Oval Office. Bessent, Lutnick, and Cairncross have all taken their positions directly to her. The source article notes that, as on other administration issues, Wiles’ decision generally becomes final.
Why the AI Policy Split Matters
The debate brings together several overlapping questions:
- Should the US restrict Chinese AI labs more aggressively?
- Should American labs be encouraged to release open-weight models?
- How should Washington respond to distillation?
- Which risks matter more: slowing innovation, losing ground to China, or allowing powerful AI systems to be misused?
The Treasury Department’s concern reaches beyond competition with China. Sources say the department has focused on the risk of AI going rogue, including the possibility that a model could inadvertently close out trading positions or access sensitive information at banks like Goldman Sachs. The fear is that such behavior could crash the economy.
Luke Pettit, Assistant Secretary for Financial Institutions at the Treasury, has become Bessent’s AI point person at the department. Pettit, previously a senior policy adviser in the US Senate, is working with the Treasury’s chief information officer and former DOGE member Sam Corcos.
The administration’s AI policy is therefore being formed through a mix of security concerns, trade pressure, industry contact, and political review. The common goal is to keep the US ahead. The unresolved question is how much restriction, openness, and confrontation with Chinese labs that goal requires.