AI & Elections / Synthetic Political Media

Current landscape

At the federal level, no election-specific deepfake statute exists. The FEC declined to open formal rulemaking in 2024, instead issuing a non-binding interpretive rule stating that existing fraudulent-misrepresentation law already covers deceptive AI content: case-by-case enforcement only, no new prohibition. The FCC has two separate, unequal tracks: a 2024 declaratory ruling that AI-cloned robocall voices violate the TCPA is finalized and in force (issued directly in response to the January 2024 New Hampshire Biden robocall incident), while a broader rulemaking proposing AI-disclosure requirements for broadcast political ads passed on a party-line vote and appears to have stalled under the current FCC chairman's deregulatory posture. Congress has reintroduced federal deepfake-election bills repeatedly without passage.

Into that federal gap, roughly 30 states have passed election-deepfake laws as of mid-2026 (the exact count varies by tracker; cite a live source rather than a fixed number). Most require disclosure/labeling; a minority impose outright bans near elections; Colorado and New York grant candidates a private right of action.

The First Amendment record

This is where the tension stops being theoretical. California's AB 2839 was struck down on summary judgment in August 2025 as unconstitutional compelled, viewpoint-based speech, citing the principle that even knowing falsehoods are generally protected absent a specific, narrow harm exception; California's appeal to the Ninth Circuit remains unresolved. Hawaii's equivalent law was permanently enjoined on similar grounds. Minnesota's law survived a preliminary-injunction challenge at the Eighth Circuit, but on standing and timing grounds, not on the merits, so it hasn't been validated as constitutional either; it's simply not yet been struck down. The pattern across all three cases points the same direction: broad "materially deceptive content" categories are losing, and laws narrowly targeted at verifiable factual claims (like voting mechanics) have a meaningfully better constitutional track record.

Incidents, and how little enforcement they produced

The New Hampshire Biden robocall remains the clearest case study, and its outcome is a caution against assuming these laws have teeth: the FCC's $1M settlement with the carrier was paid, but its $6M fine against the consultant who orchestrated the calls reportedly remains unpaid and disputed; that consultant was acquitted on all 22 associated state criminal counts in June 2025; a separate $22,500 civil default judgment against him is reportedly being defied. A March 2026 AI deepfake ad targeting a Texas Senate candidate drew public criticism but no legal consequence at all. A recent policy-research review found no confirmed prosecutions under any state deepfake law as of mid-2026, describing the laws as "largely symbolic" in practice so far.

International contrast

The EU's AI Act includes a deepfake-labeling obligation (Article 50) that does not become legally enforceable until August 2026, meaning it has not yet applied to any election as of this writing. South Korea runs the most aggressively enforced regime, with a 90-day pre-election ban carrying criminal penalties and its first fines issued in December 2025 (a constitutional challenge is now pending). Brazil's electoral court imposes an outright ban with candidacy-annulment risk. These are meaningfully different regulatory postures than the U.S.'s current patchwork, and worth watching as comparative evidence of what stronger enforcement produces or fails to produce.

Trade-offs and opposing views

Civil-liberties advocates argue that any compelled-disclosure or content-based deepfake law is constitutionally doomed given the Kohls/Hawaii track record, regardless of how it's scoped. The party's narrower proposal, targeting voting-mechanics falsehoods specifically, is a direct response: mechanics-focused election law (what counts as protected factual, non-viewpoint information about how and when to vote) has a long history of surviving scrutiny that broad "deceptive content" categories don't share. Election-integrity advocates, on the other side, argue a narrow rule leaves too much exposed: a character-attack deepfake released the night before an election isn't touched by a voting-mechanics rule at all. The party does not claim the narrow legal rule solves the whole problem. It's paired deliberately with Proposal 4 (provenance standards) precisely because the legal remedy alone is insufficient: an acknowledged gap, not a rhetorical dodge.