The Lobbyists Show Up Every Quarter

The Trump administration is not planning an outright ban on Chinese AI — it is planning everything that feels like one. On July 20, Axios reported that Kimi K3's launch reignited a secret internal fight over how to kneecap Chinese open-weight models without triggering the legal and political blowback a formal prohibition would invite.

What's new:

• Commerce last year weighed adding multiple Chinese AI labs to the Entity List, Axios reported, citing administration sources — the same designation that already covers Zhipu AI since January 2025.
• The White House considered an executive order requiring U.S. companies to guarantee security and accept liability if they host Chinese models, per Axios.
• Commerce circulated draft rules last summer leveraging supply-chain authorities to target Chinese open-source models, another source told Axios.
• The NSA and White House cyber office also weighed a public advisory discouraging U.S. firms from using Chinese AI labs.
• None of these became formal policy. All remain live options.

Why no outright ban:

• Open-weight models cannot be recalled once published. Kimi K3's full 2.8-trillion-parameter weights ship July 27.
• Brookings fellow Kyle Chan told CNBC in July that banning China's open-source models is "ultimately impossible" because weights are "available freely on the internet" — and may raise First Amendment issues.
• Chinese models already account for 45–46% of tokens routed through OpenRouter, CNBC reported July 8 — up from 4.5% in early 2025.
• A Citi note cited by industry press pegs leading Chinese models at roughly 18 cents per million tokens versus about $4 for U.S. frontier alternatives.

The playbook instead:

• Axios quoted one source familiar with government discussions: "What's actually happening is slower and more durable" — procurement rules, Entity List threats, and public pressure on U.S. companies still using Chinese models.
• Another source described highlighting potential backdoors and governance gaps — an offensive push to build a more competitive U.S. open-source ecosystem rather than a courtroom fight over downloaded weights.
• Axios reported the approaches would have a "chilling effect" on Chinese open-source tech and the American startups routing to it — without ever signing a ban.

Who's lobbying:

• Axios cited a source close to the administration saying leading AI labs or their allies approach the White House every three to five months with proposals to ban open-source models.
• On Sunday, White House AI adviser David Sacks wrote on X: "We are at a critical inflection point in AI policy. The leading closed labs, already a duopoly in terms of AI model revenue, want the government to eliminate their open-source competition."
• Altimeter's Brad Gerstner amplified the post, warning that "reg capture is bad in all forms" and that America needs "robust competition between open source & closed frontier labs."

The open fight inside the tent:

• OpenAI strategic futures head Dean Ball — who helped draft the administration's AI Action Plan before joining the lab — suggested Washington's "best strategy" may be creating regulatory risk around Chinese models, Axios reported July 19.
• Sacks asked whether Ball was "confessing to a regulatory capture strategy." Ball later clarified he was not advocating "ill-justified soft-law discouragement of Chinese AI."
• Defense Under Secretary Emil Michael fired back Sunday on X: "Dean Ball has perhaps the biggest gap between actual IQ and his own perceived IQ of anyone in the industry (about 40 points)."
• Michael argued agency bans on Chinese AI for federal employees happened through a "democratic process," not a "Deep State scheme."

What Commerce already sat on:

• Reuters reported June 16 that DeepSeek, CXMT, and more than 100 other firms were approved by an interagency committee for Entity List addition in 2025 — but Commerce has not published the designations.
• At least 75 Chinese entities in semiconductors and AI modeling cleared the committee and were slated for blacklisting, Reuters reported.
• Sources told Reuters that Under Secretary Jeffrey Kessler has sought to avoid listing Chinese parties since late 2025 for fear of escalating tensions with Beijing.
• Kimi K3 may force Commerce's hand. Axios reported the interagency fight is live again.

What enterprises should watch:

• Federal procurement walls are already real: the FY2026 NDAA orders DoD and intelligence contractors to purge DeepSeek; Alibaba's Pentagon 1260H ban took effect June 30.
• House Homeland Security and the China Select Committee are actively probing Cursor-maker Anysphere and Airbnb over Chinese model use in production systems.
• The State Department stated July 8 that Chinese AI in American enterprise systems "raises serious concerns" about censorship and CCP-aligned outputs.
• Self-hosted open weights on U.S. servers are harder to reach than API calls routed through Chinese clouds — but compliance teams will not wait for case law.

Convina's view: Washington learned what Beijing already knows: you cannot un-publish a model. The fight moved from statutes to atmospherics — Entity List threats held in a drawer, liability rules drafted but unsigned, and a quarterly lobbying rhythm to turn fear into procurement policy. Enterprises treating Chinese open weights as a cost hack should price in that FUD is now the enforcement mechanism, and the closed labs Sacks named are not the only ones writing the script.