US Congress Targets Chinese AI with ‘No Adversarial AI Act

Washington has fired another shot in the escalating technology standoff with Beijing. A bipartisan group of lawmakers led by Rep. John Moolenaar (R-MI) and Rep. Raja Krishnamoorthi (D-IL) introduced the No Adversarial AI Act, a bill that would bar every U.S. executive-branch agency from buying, licensing, or even running artificial-intelligence models developed in China, Russia, Iran or North Korea. The proposal creates a rolling blacklist, overseen by the Federal Acquisition Security Council, that procurement officers must consult before they sign a software contract; any carve-outs would require explicit sign-off from the Office of Management and Budget—or, in rare cases, from Congress itself. All of these come as Chinese Telcos along with partners announced sweeping AI innovation in telecoms and other every day use at last week’s concluded MWC Shanghai 2025. 

Capitol Hill’s patience snapped after security researchers linked code inside DeepSeek, a fast-growing Chinese generative-AI firm, to China Mobile infrastructure and flagged data flows that could expose user credentials. DeepSeek was already under scrutiny for touting a ChatGPT-class model trained at a fraction of the cost, prompting several U.S. companies and agencies to ban the software outright. Lawmakers say the episode proved how easily an alluring AI demo can mask entanglements with China’s military or intelligence apparatus. 

If passed, the law would force federal CIOs to audit everything from customer-service chatbots on Social Security websites to computer-vision models that analyse satellite imagery for the Pentagon. Any AI system with roots in the named adversary nations—whether the core weights were trained there or merely fine-tuned—would have to be ripped out or receive an exemption. Contractors would also be on the hook: deliverables that rely on Chinese code assistants or Russian vision libraries could be rejected outright.

Congress has fenced off Chinese tech before. Section 889 of the 2019 NDAA banned Huawei and ZTE gear from federal networks, and the Secure Equipment Act choked off FCC licences for Chinese telecom hardware. But AI is a different beast: models are often open-sourced, checkpoint weights hop across GitHub forks in seconds, and smaller firms fine-tune overseas models without always realising it. Open-source advocates worry the bill’s wording—“developed, in whole or in part, in a foreign adversary country”—could rope in community checkpoints that merely passed through a Beijing-hosted server. That ambiguity leaves enforcement to the yet-untested judgement of the Federal Acquisition Security Council.

Ripple effects beyond D.C.

• Supply-chain fragmentation: Allies from Canada to the EU could feel pressure to copy-and-paste the restriction, splintering AI ecosystems much like 5G supply chains fractured after Huawei bans.
• Enterprise whiplash: Federal contractors relying on code-generation tools that quietly ingest Chinese-trained embeddings may need costly rewrites or face non-compliance penalties.
• GPU diplomacy: The bill lands as Washington already limits Nvidia’s high-end AI chips from shipping to China; banning Chinese models on U.S. soil completes a tit-for-tat cycle that could accelerate Beijing’s push for domestic silicon self-sufficiency.

The House Homeland Security and Oversight committees are expected to fast-track hearings in July. Companion legislation will be shepherded through the Senate by Sen. Rick Scott (R-FL) and Sen. Gary Peters (D-MI). Given the TikTok divest-or-ban precedent and election-year hawkishness, analysts give the bill a strong shot at clearing both chambers—though industry lobbyists will fight for clearer definitions of “Chinese AI” and carve-outs for open-source research.

If TikTok was Washington’s opening salvo, the No Adversarial AI Act aims straight at the algorithmic heart of next-gen software. In practical terms, it could reshape federal IT procurement for a decade, force a wave of internal AI audits, and bake geostrategic lines into the very code that runs HR portals and drone-analysis stacks. In geopolitical terms, it marks a shift from targeting physical infrastructure to targeting the invisible logic that increasingly drives it—setting up a new, and potentially wider, tech-cold-war front.