Montana Deepfake Law Threatens Free Political Expression

A federal judge recently struck down Montana’s AI deepfake election law as unconstitutional, ruling it infringes on the political speech that sits at the heart of First Amendment protections.

U.S. District Judge Susan Watters handed a victory to a conservative political action committee accused of violating Montana’s new election law regulating AI-generated content in campaign materials. The ruling, issued this week, found that the Montana law crosses constitutional boundaries by treating identical content differently based on whether the message supports or attacks a candidate. This distinction, Watters concluded, amounts to viewpoint-based discrimination that the First Amendment simply won’t tolerate.

The case emerged from mailers sent before Montana’s June primary elections. The PAC, with former Republican legislator Dan Bartel listed as treasurer, distributed materials that used AI to alter images of Republican candidates, showing them with Pride flags and buttons. Three Republican lawmakers-Rep. Eric Albus from the Hi-Line, former legislator Jennifer Carlson running for a state House seat near Bozeman, and Rep. Llew Jones in a state senate race-filed complaints with the Commissioner of Political Practices.

Montana passed its AI deepfake election law in 2025 as lawmakers nationwide scrambled to address artificial intelligence’s growing capacity to create realistic-looking images, videos, and audio of people who never said or did what the content depicts. The bipartisan legislation aimed to curb deceptive deepfakes by regulating their use within 60 days of an election.

The statute required disclaimers on ads using AI-generated content and prohibited distribution of deepfakes intended to injure a candidate’s reputation or deceive voters. On paper, this sounds reasonable. AI technology has advanced to the point where distinguishing real footage from fabricated content has become difficult for average voters. The concern that bad actors might weaponize this technology to spread misinformation during critical election periods isn’t paranoid speculation.

But Judge Watters identified a fatal flaw in how the law operated. The statute treated “identically deceptive media differently based solely on whether its message is weaponized or deployed to elevate a candidate.” In other words, if someone created an AI-altered image praising a candidate, the law applied one standard. If that same person created an identical deepfake attacking the candidate, different rules kicked in. This distinction based on the viewpoint expressed made the law “presumptively unconstitutional” under First Amendment doctrine.

A federal judge blocked Montana from enforcing its AI deepfake election law against a conservative political action committee, finding that the statute likely discriminates based on viewpoint and burdens highly protected political speech. The ruling does not invalidate the law statewide because Trump v. CASA limits federal injunctions to the named parties, leaving states to reconsider how disclosure and anti-deepfake rules can protect elections without violating the First Amendment.

Watters emphasized that political speech receives the highest level of First Amendment protection. Courts have consistently held that the government faces an extremely heavy burden when attempting to restrict what people can say about elections and candidates. The rationale makes sense: democracy depends on robust public debate, even when that debate gets messy, misleading, or downright false.

The Hamilton Lincoln Law Institute, which represented Bartel and the PAC, argued that false political speech still falls under constitutional protection. Adam Schulman, a senior attorney at the institute, called the ruling “an important victory for political speech and the First Amendment,” adding that “the government cannot censor speech just because that speech uses AI technology.”

This position draws on established First Amendment principles. Courts have long recognized that elections involve noisy, chaotic discourse where truth and falsehood mix freely. Voters hear exaggerations, distortions, and outright lies from all sides. The constitutional solution isn’t government censorship but more speech-counter-arguments, fact-checking, and public debate that exposes falsehoods.

The Hamilton Lincoln Law Institute’s brief in the case put it bluntly: “Elections have always been a noisy ruckus of misinformation. Because the Act regulates political speech for a reason other than avoiding corruption, it is unconstitutional.” Montana’s law didn’t target the narrow category of speech related to corruption that courts sometimes allow governments to regulate. Instead, it tried to police false or misleading campaign content more broadly, triggering strict constitutional scrutiny.

Despite declaring Montana’s law problematic, Watters issued a narrow ruling that only benefits Bartel and his PAC. The decision doesn’t strike down the statute statewide or prevent enforcement against other individuals or groups. This limited scope reflects recent Supreme Court guidance on how federal judges should handle preliminary injunctions.

The Trump v. CASA decision from June 2025 dramatically reshaped injunction practice in federal courts. That case involved a presidential executive order on immigration, and the Supreme Court’s conservative majority ruled that district courts should decide cases only for the specific parties before them, not issue universal injunctions benefiting the general public.

The Court held that the Judiciary Act of 1789 doesn’t give federal courts authority to impose universal injunctions. This represents a major shift in judicial power. For years, federal judges routinely issued nationwide injunctions blocking laws or policies from taking effect anywhere in the country. Trump v. CASA ended that practice, requiring judges to limit relief to the named plaintiffs in each case.

Watters cited Trump v. CASA extensively in her decision. Montana’s lawyers defending the AI law actually requested this limited approach, asking Watters to make any injunction “no more burdensome than necessary” and pointing to the Supreme Court precedent. The plaintiffs had asked for broader relief that would’ve effectively nullified the law statewide. Watters wrote that she lacked authority under the Judiciary Act of 1789 to issue a universal injunction, and that equitable relief must be “limited to the named plaintiffs.”

This creates an odd situation where Montana’s law remains on the books and enforceable against everyone except Bartel and his PAC. Other political groups using AI-generated content in Montana elections still risk prosecution. They’d need to file their own lawsuits and win their own preliminary injunctions to get similar protection.

Montana’s legitimate concern about convincing election disinformation does not excuse a law that changes its treatment of identical synthetic media based on whom the message helps or hurts. The more durable approach is to target concrete harms and require clear, content-neutral disclosure rather than giving government officials authority to decide which political distortions may be published.

The narrowness of Watters’s ruling doesn’t erase the constitutional problems she identified. If Montana’s law engages in viewpoint discrimination and violates the First Amendment as applied to Bartel’s PAC, it presumably suffers from the same defects when applied to anyone else.

Viewpoint discrimination occurs when the government restricts speech based on the specific message or perspective expressed. Content-based restrictions-which regulate subject matter without targeting particular viewpoints-face strict scrutiny and rarely survive constitutional challenge. Viewpoint discrimination faces even tougher standards and almost never passes muster.

Watters found that Montana’s law crossed this line by treating pro-candidate and anti-candidate deepfakes differently. The technology and deceptiveness might be identical, but the law’s application hinged on whether the content helped or hurt a candidate’s campaign. That’s textbook viewpoint discrimination.

The decision raises questions about how states can regulate AI deepfakes in elections without running afoul of the First Amendment. Twenty-nine states have AI election deepfake laws on the books as of mid-2026, according to background research. These statutes vary widely. Some require disclosures about AI usage in political ads. Others ban political deepfakes within certain windows before elections. California and Hawaii saw their AI deepfake laws struck down on First Amendment grounds, leading those states to pursue alternative approaches.

The patchwork of state regulations reflects genuine uncertainty about how to balance legitimate concerns about deceptive AI content with constitutional protections for political speech. Courts reviewing these laws haven’t reached consensus either. Some judges find narrowly tailored disclosure requirements acceptable. Others see any restriction on AI-generated political content as impermissible censorship.

Defenders of AI deepfake laws argue that synthetic media presents novel challenges that justify new regulations. Traditional political lies involved verbal claims or crude photo manipulation that audiences could often detect. Modern AI tools create convincing fake videos showing candidates saying things they never said, appearing in places they never visited, or engaging in conduct that never occurred. The realism makes detection difficult even for sophisticated viewers.

But the counterargument-embraced by Judge Watters and the Hamilton Lincoln Law Institute-maintains that the First Amendment doesn’t distinguish between technologies. Political speech enjoys robust protection whether delivered through newspapers, radio, television, social media, or AI-generated content. The government can’t justify censorship simply because new tools make lying easier or more convincing.

This perspective treats AI as just another communications technology. People have always had tools for creating false or misleading political content. The printing press made mass distribution of propaganda possible. Photo editing software enabled image manipulation. Video editing allows selective cuts that distort meaning. AI represents another step in this technological evolution, but the constitutional principles remain constant.

The Hamilton Lincoln Law Institute stated this position clearly: “There is absolutely no reason why the government should be telling private citizens what they can and cannot say about their political opponents.” The technology used to craft the message doesn’t change the fundamental free speech analysis.

Even before Watters’s ruling, enforcing AI deepfake laws proved difficult. The 2026 midterm elections provided the first major test of state regulations, and the results showed spotty compliance and uneven enforcement. Some instances involved labeled satire where creators disclosed AI usage. Others featured unlabeled hostile or deceptive deepfakes that sparked investigations in states like Oregon and Michigan.

States took different approaches to timing restrictions. Minnesota and Texas prohibit distribution of deepfakes close to elections. Maryland bans them year-round. But identifying violations, gathering evidence, and taking enforcement action within compressed election timeframes presents practical obstacles. By the time authorities investigate complaints and determine whether violations occurred, elections have often passed.

Montana’s law allowed candidates to file complaints with the Commissioner of Political Practices, which Albus, Carlson, and Jones did after seeing Bartel’s mailers depicting them with Pride imagery. The complaints triggered the enforcement process that led to the lawsuit and eventual court challenge. This reactive model depends on targets of deepfakes noticing the content and filing timely complaints.

The limited reach of Watters’s injunction complicates enforcement further. Montana officials can’t apply the law to Bartel’s PAC but must enforce it against everyone else. This creates equal protection concerns. If two political groups distribute similar AI-generated content, why should one face prosecution while the other gets immunity? The practical answer-one group successfully challenged the law in court while the other didn’t-may satisfy legal process requirements but produces arbitrary results.

The Montana ruling doesn’t resolve the larger debate over regulating AI in elections. Watters issued a preliminary injunction, finding that Bartel and his PAC are likely to succeed on the merits of their constitutional claim. The case could proceed to a full trial, though the limited scope of relief reduces incentives for extensive litigation.

Montana could appeal Watters’s decision, though an appeal would only affect whether Bartel gets relief, not whether the law stands more broadly. The state could also revise its statute to address the viewpoint discrimination problem Watters identified, perhaps by crafting content-neutral disclosure requirements that don’t hinge on whether AI content helps or hurts candidates.

Other states watching Montana’s experience might reconsider their own AI deepfake laws. Some may decide that disclosure requirements survive constitutional scrutiny better than outright bans. Others might conclude that any regulation targeting AI-generated political content risks First Amendment challenges and abandon the effort.

At the federal level, no comprehensive standard for election AI deepfakes exists. While the Take It Down Act addresses nonconsensual intimate imagery including AI-generated content, election deepfake legislation remains pending in Congress. States have moved faster than federal lawmakers, but the resulting patchwork creates confusion and inconsistency.

The Trump v. CASA decision limiting universal injunctions may actually increase pressure for federal standards. When district courts could issue nationwide injunctions blocking problematic laws, a single successful challenge provided relief broadly. Now, each affected party must bring separate litigation and win individual injunctions. This fragmented approach makes uniform constitutional protection harder to achieve and may push stakeholders toward seeking legislative solutions that avoid constitutional problems.

Montana’s failed deepfake law illustrates the fundamental tension between protecting election integrity and preserving free speech. Nobody disputes that AI technology can create convincing false content. Nobody denies that such content could mislead voters and distort democratic processes. The question is whether these real risks justify government regulation of political speech.

Judge Watters answered no, at least for Montana’s specific statutory approach. The law’s viewpoint-based distinctions made it constitutionally unacceptable regardless of the legitimate interests it served. This doesn’t mean all AI election regulations will fail constitutional review, but it confirms that courts will apply rigorous First Amendment scrutiny to laws restricting political speech based on technology used or viewpoint expressed.

The decision reinforces that political speech-even false, misleading, or AI-generated political speech-receives maximum constitutional protection. Lawmakers uncomfortable with this reality face difficult choices. They can craft narrower regulations focused on disclosure rather than prohibition. They can target specific harms like fraud or defamation through existing legal frameworks. Or they can accept that AI-generated political content, like other controversial speech, will remain largely unregulated despite potential risks.

For defenders of free expression, the Montana ruling represents an important victory against government overreach. For those worried about AI-fueled disinformation, it highlights the constitutional obstacles to regulation and the need for solutions that don’t rely on restricting speech. The debate will continue as technology evolves and courts refine First Amendment doctrine for the AI age.