Don't California My Arizona: Why the First Amendment Needs Arizona’s Playbook on AI Political Ads
By: Linley Wilson and Claire Woynerowski
During the 2026 Los Angeles mayoral race, viral AI-generated videos of candidate Spencer Pratt depicted him as a lightsaber-wielding Jedi and Batman confronting a Joker-inspired Mayor Karen Bass. The Pratt videos illustrate that how states regulate AI in political ads matters as much as whether they regulate at all.
Unsurprisingly, California and Arizona have taken very different approaches. California imposed broad, content-based bans that courts struck down, while Arizona crafted narrow disclosure rules and civil remedies that protect election integrity without chilling political speech. Their contrasting results offer a roadmap for other states to legislate within constitutional limits.
California’s Overreach and Its Consequences
California tried to address AI in political advertising through a series of three bills—AB 2355, AB 2839, and AB 2655—but went too far. AB 2355 requires political committees’ ads containing AI-generated content to include a disclaimer: “Ad generated or substantially altered using artificial intelligence.” Importantly, AB 2355 applies only to ads created by committees defined under California law, not to content created by individual citizens.
But AB 2839 went further, banning the knowing distribution of “materially deceptive” AI-generated election content and allowing virtually any recipient (not just the depicted candidate) to sue for damages and injunctive relief. AB 2655 required large online platforms to block “materially deceptive” political deepfakes and label certain content as inauthentic during windows before and after elections.
Both laws were invalidated in Kohls v. Bonta, 797 F. Supp. 3d 1177 (E.D. Cal. 2025). The district court found, among other things, that AB 2839 “discriminates based on content, viewpoint, and speaker and targets constitutionally protected speech.” The court also ruled that AB 2655 is preempted by Section 230 of the Communications Decency Act, a decision now on appeal in the Ninth Circuit.
The 2026 Los Angeles mayoral race highlights the real-world consequences of state legislation that infringes on constitutional rights. Spencer Pratt, the former MTV reality star turned mayoral candidate, saw his campaign boosted by AI-generated videos made by individual supporters—not the campaign itself. AB 2355 didn’t apply to these viral videos because the creators were individuals, not committees, and AB 2839—the overbroad law that would have arguably reached them—was enjoined.
The Kohls court correctly identified the constitutional infirmities in AB 2839. The bill wasn’t narrowly tailored because it swept up protected speech, including satire, parody, and political commentary, alongside genuinely fraudulent content. When a supporter creates a stylized AI video depicting a candidate as a superhero, as in the Pratt videos, that is political commentary—not fraud. The First Amendment demands narrowly drawn measures, not sweeping bans on AI-assisted political expression.
Arizona’s Constitutional Model for AI Election Law
Arizona shows that meaningful regulation of AI-generated political content can satisfy the First Amendment. Instead of banning or censoring such content, Arizona enacted two complementary laws, SB 1359 and HB 2394, built on targeted disclosure requirements and carefully limited civil remedies.
SB 1359 requires creators of “deceptive and fraudulent deepfakes” depicting political candidates to clearly disclose the content was AI-generated, but only within 90 days before an election. Penalties are modest—$10 per day for the first fifteen days and $25 per day thereafter. The law expressly exempts satire and parody.
HB 2394 gives candidates a right to sue over digital impersonation against a publisher within two years. To prevail, a plaintiff must show the impersonation was published without consent or that the publisher didn’t make clear the content was inauthentic. The primary remedy is a court declaration, not damages, and courts must rule on emergency relief within two business days.
Arizona’s framework passes constitutional muster because it is narrowly tailored. Where California sought to prohibit “materially deceptive” content—a vague category that captured satire and parody—Arizona targets only content that is both deceptive and fraudulent, and only during a 90-day pre-election window. Where California let virtually anyone sue for damages, Arizona limits claims to the depicted person and focuses on court declarations rather than monetary awards that could chill speech through litigation threats.
The result is a regulatory scheme that protects candidates from genuinely fraudulent deepfakes while leaving untouched the broad universe of AI-generated political expression—satire, commentary, creative advocacy, and the kind of fan-made videos that defined the Pratt campaign. Arizona’s approach allows for reasonable safeguards without infringing on First Amendment rights.
As more states legislate ahead of the 2026 midterms and beyond, lawmakers should learn from Arizona’s success and California’s overreach. AI is a new tool for political expression, and like the printing press and the internet before it, its use by citizens for political speech deserves the highest constitutional protection. State legislators, advocacy groups, and legal practitioners should embrace the Arizona model, challenge overbroad laws like California’s in court, and promote narrowly-tailored disclosure frameworks that respect both election integrity and the First Amendment.
