The late-September 2026 Trump–Xi summit produced neither an AI safety treaty nor a verifiable pause. What remains on the U.S. domestic board are bills that have circulated for months: the Remote Access Security Act and the MATCH Act, cited explicitly by the 17 September CNAS insight as tools to close gaps Beijing exploits.
The Remote Access Security Act (H.R. 2683) passed the House on 12 January 2026 and remains in the Senate Banking Committee. Its logic is direct: if Commerce does not currently interpret its authority to cover remote access to export-controlled chips, Congress must clarify and mandate regulations. Without that piece, an H100 in a Malaysia rack rented by API is a datacenter-sized hole.
The MATCH Act (H.R. 8170) aims to harmonize allied restrictions on semiconductor manufacturing equipment toward countries of concern, protecting U.S. AI hardware leadership. September reporting cast it as a candidate for inclusion in Senate NDAA manager amendments for FY2027, though the NDAA itself had weeks of procedural stall after summer cloture failures.
CNAS links both texts to three gaps: remote access (~1M H100-equivalents via SE Asia), adversarial distillation (Alibaba/Claude case) and ASML/Tokyo Electron servicing in China that extends DUVi life and transfers know-how. MATCH and Remote Access do not solve distillation alone—that needs antitrust safe harbors and lab sharing—but they attack compute and equipment.
Neural Pulse stresses the timing gap. Even if the Senate passes and the president signs, Commerce must promulgate specific rules. That administrative process can take months while remote access continues. Legislation is necessary and not sufficient; without operating regulations, the U.S. Code title is a statement.
The industrial context complicates policy. Allies (Netherlands, Japan) resist abrupt servicing cuts that hit revenue and China-market presence. SE Asia cloud providers argue commercial neutrality. U.S. frontier labs want abundant allied compute without leakage to adversaries. MATCH tries to square that triangle.
Critics of control expansion warn that tightening export controls only accelerates Chinese indigenization. CNAS answers that Beijing has been all-in since 2014; the goal is not to prevent forever but to lengthen the lead. SMIC yield data, DUVi volumes and model lag on ExploitBench support, per the center, that partial braking works.
For investors and CISOs, September's signal is: plan for scenarios in which remote access to Blackwell/H100 from third jurisdictions becomes explicitly controlled. Cloud contracts, tenant KYC and workload-location attestations will cease to be "nice to have."
The link to the Bessent–He Lifeng incident channel is indirect but real. Officials have said the AI dialogue does not imply relaxing chip controls. If Congress hardens Remote Access while Shenzhen discusses notifications, Washington's narrative coherence improves: talk security without giving away compute.
Neural Pulse does not predict 2026 passage. It predicts the post-summit debate will shift from "global pause?" to "who closes the remote API?" That shift is more prosaic and more decisive for the 2027 compute balance.
Coordination with the British G20 and industrial coalitions also matters: workload-identity standards and near-miss reporting on compute exfiltration can complement statute. Law without technical norms is paper; norms without authority are voluntarism.
Close: MATCH and the Remote Access Security Act are the boring names that decide whether the CNAS insight stays a PDF or becomes Code of Federal Regulations. September 2026 left them unfinished; the legislative autumn will show whether the window Nie and Remler call "narrow" is used or missed.
