California Law Restricting "Materially Deceptive" Election-Related Deepfakes Violates First Amendment

So a federal judge just held.

From today's decision by Judge John Mendez (E.D. Cal.) in Kohls v. Bonta:

AB 2839 regulates a broad spectrum of election-related content that is "materially deceptive" and permits any recipient of this content to sue for general or special damages. Cal. Elec. Code §§ 20012(b)(1), 20012(d). AB 2839 defines "materially deceptive" content as "audio or visual media that is intentionally digitally created or modified, … such that the content would falsely appear to a reasonable person to be an authentic record of the content depicted in the media." AB 2839 includes exceptions for candidates who make and share deepfake content of themselves and for satire or parody. In both these cases, the content must include a disclaimer that meets AB 2839's formatting requirements and must state that the content has been digitally manipulated….

The Court finds that AB 2839 discriminates based on content, viewpoint, and speaker and targets constitutionally protected speech.

The Court's preliminary injunction Order recognized that AB 2839 was likely unconstitutional because it was content-based. By its terms, AB 2839 prohibits "materially deceptive" (defined as content that would falsely appear to a reasonable person to be an authentic record) audio or visual communications that portray a candidate or elected official doing or saying things he or she didn't do or say and that are likely to harm a candidate's reputation or electoral prospects. The statute also punishes such altered content that depicts an "elections official" or "voting machine, ballot, voting site, or other property or equipment" that is "reasonably likely" to falsely "undermine confidence" in the outcome of an election contest. As evidenced by the statutory language, AB 2839 facially regulates based on content because the "law applies to particular speech because of the topic" —a political candidate, elected official, elections official, ballot, or voting mechanism. Moreover, it delineates acceptable and unacceptable speech based on its purported truth or falsity meaning that non-materially deceptive content is excluded. See Order at 11.

On top of the content-based distinctions, AB 2839 regulates speech based on viewpoint and speaker. The state law only punishes content that could "harm" a candidate's electoral prospects or content that could "undermine confidence" in the outcome of an election while leaving positive representations unregulated. In other words, materially deceptive content that helps a candidate or promotes confidence would not be subject to penalty under AB 2839….

Moreover, AB 2839 also engages in speaker-based discrimination because the law imposes different obligations on different speakers depending on who they are. Under AB 2839, candidates posting about themselves, broadcasters, and internet websites are subject to more lenient rules while other speakers, such as Plaintiffs, are categorically barred. Candidates and broadcasters can post "materially deceptive" content as long as they attach disclaimers. Additionally, broadcasters and internet sites are exempt from "general or special damages." AB 2839 treats different speakers dissimilarly, subjecting certain individuals to stricter rules and other speakers to more lenient rules. All together, these content, viewpoint, and speaker-based distinctions at minimum trigger strict scrutiny….

Attempting to avoid the content, viewpoint, and speaker- based problems with AB 2839, Defendants analogize the statute to narrow categories of historically recognized exceptions to the First Amendment such as defamation or fraud…. However, AB 2839 goes beyond these historical categories. For example, the statute diverges from defamation law because it proscribes content that is merely "reasonably likely" to cause harm, which is speculative and prophylactic rather than remedial or concrete. Moreover, the statute also goes beyond reputational harms to include amorphous harms to the "electoral prospects" of a candidate.

So too do AB 2839's regulations go beyond the definition of fraud because unlike fraud, AB 2839 does not require reliance or actual injury. See United States v. Alvarez (2012) (Breyer, J., concurring). California responds that falsehoods "meant to deceive viewers and manipulate voters to change their voting behavior" do cause legally cognizable harm, but intent to "deceive and manipulate" alone is not sufficient under Alvarez, which recognized that even knowing falsehoods are constitutionally protected….

Notably, the most significant manner in which AB 2839 goes beyond historically recognized exceptions to the First Amendment is by deputizing a much more expansive category of plaintiffs. Unlike defamation or other tort remedies that limit plaintiffs to persons actually harmed, the category of plaintiffs AB 2839 cognizes is almost boundless because it allows the government as well as any recipient of materially deceptive content to "seek injunctive or other equitable relief." Plus, these recipients can seek "general or special damages" and "attorney's fees and costs," even against a person who merely "republishe[s]" prohibited content. Allowing almost any person to file a complaint creates the "real risk" of malicious lawsuits that could chill protected speech. Susan B. Anthony List v. Driehaus (2014).

Rather than targeting content that procures tangible harms or materially benefits a speaker, AB 2839 attempts to stifle speech before it occurs or actually harms anyone as long as it is "reasonably likely" to do so and it allows almost anyone to act as a censorship czar….

[S]trict scrutiny is the appropriate standard for a content-based restriction that implicates political expression like AB 2839…. To withstand strict scrutiny, AB 2839 must advance a compelling state interest through the least-restrictive means possible….

While the Court acknowledges that California may have a compelling interest in protecting election integrity, the tools it deploys to achieve its interest must be the least restrictive means of achieving such goal when significant speech issues are at stake. As Plaintiffs argue, the most glaring issue with AB 2839 is that the statute is not narrowly tailored because it captures even constitutional deepfakes and all "materially deceptive content." The First Amendment does not "permit speech-restrictive measures when the state may remedy the problem by implementing or enforcing laws that do not infringe on speech." "Because restricting speech should be the government's tool of last resort, the availability of obvious less-restrictive alternatives renders a speech restriction overinclusive" and unconstitutional.

As the Court previously recognized in its preliminary injunction Order, existing statutory causes of action, including "privacy torts, copyright infringement, or defamation already provide recourse to public figures or private individuals whose reputations may be afflicted by artificially altered depictions peddled by satirists or opportunists on the internet." Indeed, several other narrower constructions might allow the statute to align with historically recognized First Amendment exceptions. For instance, California could limit AB 2839's reach to false speech that causes legally cognizable harms like false speech that actually causes voter interference, coercion, or intimidation.

California could also limit the statute's reach to factual statements that are demonstrably false like the time, date, place, or manner of voting. See generally Eugene Volokh, When are Lies Constitutionally Protected?, 4 J. Free Speech L. 685, 704–09 (2024) (contrasting lies about "election procedures"—an area where a "narrower restriction[] might pose fewer problems" with lies about election campaigns and government officials—areas that should be "categorically immune from liability").

Another narrower construction might be for California to limit potential plaintiffs to political candidates actually harmed by unprotected false speech, which would mirror defamation law more closely. Plaintiffs also suggest that California could encourage alternatives that are already working in the free market such as fact checking or counter speech.

California could even fund its own AI educational campaigns or form committees on combatting false or deceptive election content. While California's expert explains that political deepfakes are "sticky" and this type of misinformation spreads too quickly for governments to counteract it, Plaintiffs have offered evidence from their expert that shows fact-checking alternatives like "Community Notes and Grok are already … scalable solutions being adopted" in the real world.. These misinformation flagging tools crowdsource identification and labeling to educate citizens rather than relying on censorship to eradicate potentially misleading content. Thus, California provides no substantial evidence that other less restrictive means of regulating deceptive election content are not feasible or effective….

The court also concluded that plaintiffs' satires could fall within the scope of AB 2839:

Defendants agree that some satirical videos can appear to be authentic within the meaning of AB 2839. California previously represented at the preliminary injunction stage that a "voter who encountered [the Harris Parody Video] … could have concluded … that it was real." Thus, AB 2839's expansive terms capture even satire or parody videos since the law does not require that the parody in fact does fool or mislead someone.

Content need only "falsely appear … authentic" in some respect to violate the law. Since parody "imitates the characteristic style of an author or a work for comic effect or ridicule," much digitally created parody would run afoul of the law. Moreover, the State's contention that parody and satire are excepted is unpersuasive because AB 2839's safe harbor codified at Cal. Elec. Code § 20012(b)(3) imposes a disclaimer requirement on parody or satire that is independently suspect….