Detailed Analysis
A legal analysis published under the title "When an Export Order Reaches Inward" constructs a targeted constitutional challenge to hypothetical executive action that would use national security authority to ban foreign national access to a large language model — an action that, in practice, would force the model's American developer to shut the service down entirely due to the absence of technically reliable segmentation between domestic and foreign users. The article's central argument is deliberately counterintuitive: the most commonly invoked objections to such an order, namely that the AI itself has free speech rights or that its outputs are categorically protected expression, are the weakest lines of attack. The stronger constitutional vulnerabilities lie in the separation of powers, statutory limits on executive emergency authority, and the breadth of a rule that would suppress protected domestic speech as a means of reaching foreign targets.
The author methodically dismantles the AI-as-speaker argument before pivoting to more durable ground. Noting that corporate speech doctrine, from *Bellotti* through *Citizens United*, never actually dispensed with human authorship but merely located it one step removed in the choices of people acting through a corporate form, the piece concludes that a model's outputs occupy uncertain expressive territory — closer to a computed artifact than to human communication in the constitutional sense. A more defensible framing treats the AI service as a medium through which the developer company exercises genuine editorial judgment: choices about training data, reward signals, behavioral constraints, and expressive tone are dense, human, and real. The Supreme Court's 2024 decision in *Moody v. NetChoice* provides the most current doctrinal bridge, recognizing that human policy choices do not lose First Amendment protection simply because software executes them at scale. The article is candid, however, that whether this bridge holds when the software *generates* rather than merely *sorts* content remains constitutionally untested, and counsels that this argument be pleaded as one rail among several, not as the whole case.
The analysis identifies executive authority as the order's most immediately fragile point. An executive order, lacking independent constitutional moorings, must trace to a direct presidential power or a statutory delegation from Congress, neither of which clearly covers a directive to shutter a domestic software service with incidental domestic casualties. The International Emergency Economic Powers Act's Berman Amendments, which expressly carve out "informational materials" and personal communications from the President's emergency economic powers, create a direct statutory conflict if the AI's outputs qualify as information — and the same expressive analysis that might protect the company under the First Amendment simultaneously triggers this carve-out. Reinforcing this fragility, the *West Virginia v. EPA* major questions doctrine and the post-*Loper Bright* narrowing of agency deference together require clear congressional authorization before an executive agency can claim sweeping authority over a significant new domain. A directive that effectively shuts down an entire category of AI service for American citizens in order to exclude foreign nationals would have to clear both hurdles with statutory text that, as the article implies, does not plainly exist.
The piece situates this analysis within a genuine policy inflection point rather than pure legal abstraction. As AI models become infrastructural — embedded in professional, commercial, and communicative activity at scale — the practical reach of export control mechanisms extends inward in ways the existing legal architecture was not designed to handle. The technical impossibility of clean national segmentation in a cloud-delivered AI service means that foreign-targeted restrictions carry automatic domestic spillover, transforming what would ordinarily be a foreign commerce or national security question into a direct regulation of American speech and commerce. This dynamic distinguishes AI from prior export control targets like semiconductor hardware or dual-use software with discrete distribution channels. The article's contribution is to map the specific doctrinal seams — IEEPA's informational carve-out, major questions doctrine, overbreadth analysis — where that inward reach becomes legally untenable, offering practitioners and policymakers a roadmap for the litigation that current AI export policy trajectories make increasingly likely.
Read original article →