A coalition of more than 20 national-security experts and former officials, led by FDD Action, is pressing congressional leaders to bolt five artificial-intelligence export-control measures onto the must-pass FY2027 National Defense Authorization Act, warning that the U.S. lead in advanced computing is real but fragile. The centerpiece of the June 17, 2026 appeal is the bipartisan AI OVERWATCH Act (H.R. 6875), which would require the federal government to treat exports of the most advanced AI chips like arms sales subject to mandatory congressional notice and review.
Background
For two years the federal government has tried to choke off the flow of cutting-edge AI accelerators to the People's Republic of China through a patchwork of Bureau of Industry and Security rules. Critics across both parties argue that an administrative approach, revisable at any time and litigated case by case, is too brittle to govern a technology now central to military command-and-control, intelligence analysis and weapons design. The FDD Action coalition wants Congress to write the hardest lines into statute, where they cannot be quietly relaxed.
The signatories framed the stakes in stark terms. "The United States and its allies, not the CCP, should shape the future of AI," their June 17 letter to congressional leaders said. "America's lead is real but fragile." By attaching the provisions to the annual defense-policy bill, advocates aim to ride one of the few pieces of legislation that reliably becomes law each year, sidestepping the gridlock that has stalled standalone technology measures.
The push does not start from scratch. The AI OVERWATCH Act was introduced December 18, 2025 by House Foreign Affairs Committee Chairman Brian Mast, and the committee advanced it on January 21, 2026 by a 42–2 vote. Senator Jim Banks introduced a Senate companion, S. 4456, on April 30, 2026. That committee momentum is what makes the NDAA vehicle plausible: a bill that already cleared a full committee with overwhelming bipartisan support is far easier to fold into a larger package than untested language.
Key Details
The AI OVERWATCH Act would require notification to Congress and a 30-day review period before the executive branch could approve exports of advanced AI chips to a defined set of high-risk actors: China, Cuba, Iran, North Korea, Russia and Venezuela. During that window, Congress could block a sale through a joint resolution of disapproval, mirroring the procedure that governs major foreign military sales of conventional arms. The bill would also statutorily bar exports of Nvidia Blackwell-class chips to China for at least two years, converting what is now a regulatory judgment into a fixed legal prohibition.
The coalition is urging four additional measures alongside it. The Chip Security Act (H.R. 3447 / S. 1705) would mandate hardware location verification on advanced chips and require anti-smuggling reporting, an attempt to make diverted hardware trackable after it leaves a U.S. loading dock. The Remote Access Security Act (H.R. 2683 / S. 3519) would restrict PRC-affiliated access to U.S. AI compute delivered through the cloud, layering know-your-customer controls onto a channel that physical export rules do not reach. The MATCH Act (H.R. 8170 / S. 4281) would align allied controls on deep-ultraviolet lithography equipment, closing gaps between U.S. restrictions and those of partner governments. Finally, the Deterring American AI Model Theft Act (H.R. 8283) would impose sanctions for model-distillation and intellectual-property-theft attacks aimed at extracting the capabilities of leading American models.
Taken together, the package targets every layer of the AI stack that adversaries can exploit: the chips themselves, the equipment that fabricates them, the cloud that rents their output, and the trained models that run on them. The unifying move is institutional. Each bill, in different ways, would shift discretion away from executive-branch regulators and toward Congress, hard statutory bars, or verifiable compliance obligations imposed on industry.
What It Means for Contractors
For the defense-AI supply chain, the proposals would harden into law a set of obligations that companies currently navigate as shifting agency guidance. Chipmakers and the systems integrators who build AI hardware into defense platforms would face a congressional review gate on their highest-end exports, lengthening deal timelines and introducing political risk that cannot be resolved through a licensing officer. A two-year statutory bar on Blackwell-class exports to China would remove a commercial market that some vendors have lobbied to preserve, and it would do so in a way no future administration could reverse without new legislation.
Hardware location verification and anti-smuggling reporting under the Chip Security Act would land directly on manufacturers and downstream integrators, who would need to engineer tracking into products and stand up reporting pipelines to demonstrate that fielded chips remain where they are licensed to operate. Cloud providers and any contractor reselling AI compute would face know-your-customer screening obligations under the Remote Access Security Act, a compliance burden familiar to financial firms but new to much of the technology sector. Vendors selling into both commercial and federal markets would have to reconcile one compliance posture across both, since the controls attach to the technology rather than the customer.
The model-theft provision signals that the government increasingly views trained models as exportable, protectable assets, which has implications for how contractors document, secure and report on the AI systems they deliver to agencies. None of this is law yet; the measures remain proposals competing for space in an NDAA that will be negotiated through the fall. But the committee vote behind the centerpiece bill, the bipartisan sponsorship and the breadth of the coalition suggest the direction of travel. Contractors whose roadmaps assume continued access to high-end Chinese markets, or who have not built export-verification and KYC controls into their AI offerings, would be prudent to model a world in which Congress, not a single agency, holds the final say over where the most capable American AI hardware and models are allowed to go.
