Feilian Huang and Yujia Liu have posted The First AI-Saturated Election: Disclosure, Detection, and the Regulatory Limits of Generative Political Persuasion on SSRN. Here is the abstract:
The campaign cycle preceding the November 3, 2026 United States midterm elections is the first in which generative artificial intelligence has functioned as a routine instrument of political persuasion: party committees have aired synthetic videos of opposing candidates, manipulated campaign mailers have reached voters’ mailboxes, and foreign actors have seeded AI-generated disinformation through fabricated broadcast footage. Legislatures responded with unprecedented speed: as of September 2026, thirty-three states have enacted statutes regulating AI-generated election content, most built on mandatory disclosure, a minority on outright prohibition. Yet the regulatory edifice is already cracking. By September 2026, federal district courts had permanently enjoined California’s distributor-liability statute and its platform-takedown mandate (rulings now on appeal), struck down Hawaii’s criminal deepfake law, and preliminarily blocked enforcement of Montana’s deepfake statute against its challengers on viewpoint-discrimination grounds. This paper argues that state regulation of generative political persuasion faces two independent constraints that together make speaker- and content-focused regulation unworkable. First, constitutional: the First Amendment doctrines of strict scrutiny for content-based political speech, compelled-speech limits, and viewpoint-discrimination prohibitions doom most existing statutes. Second, technical: watermarking, AI-text detection, and provenance infrastructure (C2PA) cannot reliably identify AI-generated election content at scale, and each fails precisely where enforcement needs it most. The paper then proposes a regulatory framework that shifts from speaker-focused mandates to resilience signals and imposing duties on distributors and intermediaries. We also identify the constitutional exposure that such distributor duties themselves carry, and narrow the proposal accordingly. The framework offers a post-midterm legislative agenda grounded in what the technology can actually do, rather than what disclosure statutes pretend it can.
Recommended!
To receive new posts from Legal Theory Blog by email, get a free subscription to Legal Theory Stack.
Lawrence Solum