Artificial intelligence can put words in a candidate's mouth, manufacture a video that never happened, and send the fake around the world before breakfast.
Montana decided it needed a law.
Then lawmakers wrote criticism into the definition of the offense.
Senate Bill 25 defines a political "deepfake" as AI-generated or synthetic content depicting a candidate or political party "with the intent to injure the reputation" of the candidate or party or otherwise deceive a voter. During the 60 days before voting begins, political deepfakes generally require a more prominent disclosure.
The state was trying to solve a real problem. Imagine a convincing AI video appearing two days before an election showing a candidate saying something he never said. By the time anyone proves it's fake, thousands of people may have already seen it.
The government still has to obey the First Amendment when it writes the solution.
U.S. District Judge Susan Watters concluded this week that Montana's law likely crosses the line because its definition specifically singles out synthetic speech intended to damage someone's reputation.
A flattering fake about Candidate Smith and a damaging fake about Candidate Smith can both be false, maybe something akin to a Schrödinger's deepfake.
But under part of Montana's definition, the damaging one receives different treatment precisely because it attacks Smith.
Watters found the challengers likely to succeed with their argument that the law discriminates based on viewpoint.
From Reuters:
"Accordingly, because the Act treats identically deceptive media differently based solely on whether its message is weaponized to damage or deployed to elevate a candidate, the Court finds that Plaintiffs are likely to succeed in establishing the Act constitutes viewpoint-based discrimination," Watters wrote.
She issued a preliminary injunction, opens new tab that would prevent prosecutors, as well as the state agency that monitors and enforces campaign finance and disclosure laws, the Montana Commissioner of Political Practices, from investigating or prosecuting Bartel's group for violating the law.
The lawsuit came from Accountability in State Government, a conservative political committee led by former Montana legislator Dan Bartel. The group distributed altered campaign mailers depicting candidates holding pride flags and wearing preferred-pronoun pins.
Three complaints followed.
Montana's Commissioner of Political Practices eventually dismissed them because the material qualified for the law's satire exception. Bartel's group kept fighting, arguing future political speech could still expose it to investigation or punishment.
Montana argued the law also reaches material intended to "otherwise deceive a voter," language that isn't limited to negative speech.
Recommended: AI Regulation: If Bill Gates Is for It, I’m Against It. But Is It That Simple?
Watters wasn't persuaded. The state had still written a separate reputation-injury standard into the definition. A law aimed purely at deception would've been a different constitutional question.
Political speech has always been rough. Candidates have been mocked, caricatured, misquoted, distorted, and occasionally outright lied about since long before anyone plugged in a computer.
AI changes the speed and quality of the deception.
It doesn't change the First Amendment.
Watters didn't erase Montana's law for everyone. She issued a preliminary injunction protecting Bartel and his group while the case continues. She declined to impose a statewide injunction, citing the Supreme Court's recent limits on universal injunctions.
So this fight is far from finished.
More than 30 states have enacted laws addressing political deepfakes, and lawmakers everywhere face the same problem. Technology can now manufacture remarkably convincing political fiction.
States can try to address fraud, require disclosures, and establish rules that survive constitutional review.
What they can't easily do is create one rule for synthetic political praise and another for synthetic political criticism.
Montana wanted to police AI deception.
A federal judge just reminded the state that the government doesn't get to police the direction of political speech while doing it.