The First Amendment is the constraint shaping every state attempt to regulate AI deepfakes in political advertising. Courts have permanently enjoined the AI election deepfake laws of both California and Hawaii on free speech grounds, and the resulting legal vacuum is now a live variable for voters heading into the midterms, with 29 other states filling it differently.
How the 29-state patchwork actually works
The variation runs deep. Minnesota and Texas restrict political deepfakes only within a defined window before an election. Maryland's prohibition runs year-round. States outside that group generally require disclosure when AI is used in political ads, and the specificity of those disclosures differs: Colorado and Utah mandate the deepfake's creator, its creation date, and a description of how it was edited, according to the National Conference of State Legislatures.
That granularity matters because enforcement hangs on it. A generic disclosure requirement is easy to satisfy with boilerplate. Named-creator mandates tie the content to an accountable party.
California and Hawaii redirect after injunctions
Both states are pursuing secondary strategies that sidestep direct regulation of election AI deepfakes. Part of California's AI Transparency Act took effect on August 3, shifting the regulatory lever from election-specific bans to broader disclosure obligations on AI-generated images. Hawaii passed a law prohibiting deepfakes in ads without the subject's consent, an approach rooted in personal rights rather than electoral integrity.
Neither replacement regime is as direct as the original laws. The workarounds reflect what survives judicial scrutiny, not what the legislatures would have preferred.
The federal gap
Congress has no baseline standard for election AI deepfakes. The Take it Down Act, which took effect in May, marked Congress' first AI deepfakes legislation but addressed a separate issue: non-consensual intimate imagery. Advocates have already flagged inconsistent post-removal processes under that law and are predicting enforcement litigation.
House Democrats have said they plan to push legislation targeting election deepfakes specifically if they regain power next year. The current record is one AI-adjacent statute in force, zero election-specific federal rules, and an Election Day approaching on a 29-state patchwork.