Vermont attempts to regulate political AI deepfakes

Vermont lawmakers have introduced Senate Bill S.23, a proposal aimed at curbing the use of AI-generated synthetic media—commonly known as “deepfakes”—in state elections. The bill would make it a criminal offense to create or knowingly distribute digital content intended to deceive voters about a candidate’s words or actions in the 90 days leading up to an election.

The legislation attempts to overcome constitutional and logistical hurdles faced by other state restrictions on political deepfakes, as it explicitly exempts parody.

Supporters of S.23 have stressed its role in protecting the integrity of Vermont’s elections. State Sen. Ruth Hardy (D-Addison), the bill’s lead sponsor, told Vermont news site VTDigger:

“The bill is really just about making sure Vermonters can trust what they see and hear during an election. We’re not trying to stop satire or free speech. We’re trying to prevent intentionally deceptive use of AI and deepfakes that could change the outcome of an election.”

As specified in Section 3(b) of the bill: “A person shall not, within 90 days of an election … knowingly distribute or cause to be distributed … materially deceptive synthetic media of a candidate, with the intent to injure the candidate’s reputation or to deceive a voter into voting for or against the candidate.”

Violations of the bill are enforced by the attorney general and range from less than $1,000 up to $15,000, if they have previously committed a violation and intended to cause “bodily harm”.

Perhaps most importantly, the bill attempts to avoid the legal pitfalls of similar attempts in other states by exempting humorous content. Section 3(e)(2) of S.23 states: “This section does not apply to any… communication that is satire or parody.”

Parody has been a legal puzzle for those who wish to regulate deepfake content. Vermont’s S.23 closely mirrors California’s Assembly Bill 2839, which was designed to combat the use of AI-generated “deepfakes” in elections by requiring prominent disclosures on altered media. Like Vermont’s bill, the California law aimed to prevent voters from being misled by synthetic videos or audio in the lead-up to an election, but it quickly faced pushback from free speech advocates and the courts. The Foundation for Individual Rights and Expression (FIRE), a national free speech advocacy organization, criticized California’s law for its mandatory labeling of parody and satire, arguing:

“The law also requires satire and parody to be labeled, like requiring a comedian to preface every joke with an announcement he’s making a joke. That’s not funny — it’s scary.”

This critique was echoed by a federal judge who issued a preliminary injunction against most of California’s statutes. The case involves a conservative content creator, Christopher Kohls, who shares humorous political videos using AI-generated content. The court held that California’s law’s disclosure requirements rendered satire and parody unworkable.

Additionally, the question of “intent to deceive” is especially challenging in today’s media environment, where political messaging often blurs the line between parody and genuine advocacy. For example, a recent video circulating online depicted President Donald Trump and other world leaders celebrating in a glitzy, Atlantic City–style resort set in the Gaza Strip—a region currently embroiled in conflict. The video, which appeared exaggerated and outlandish, was created by a reportedly unaffiliated account, but Trump himself later shared it as a point of pride. The ambiguity of the video’s purpose—whether it was an over-the-top parody meant to mock the president’s proposal or a sincere endorsement—highlights just how difficult it is to discern intent.

Under Vermont’s bill, it would be up to a court to decide whether such content was intended to deceive or merely to satirize, creating legal uncertainty for both creators and those who share such media.

Previous state bills largely focused on publishers of content, not anyone who shares the videos. When Virginia lawmakers attempted their own restrictions on political deepfakes, the bill applied only to those who had made campaign contributions of $1,000 or more. Senate Bill 775 from state Sen. Scott Surovell, (D-Fairfax) would have required conspicuous disclosure on synthetically generated political advertisements, but it did not apply to any citizen who shared the media. The bill was ultimately vetoed by the governor, who argued that the bill’s “broad and vague approach lacks the precision necessary to ensure fair and enforceable application.”

Vermont’s attempt at regulating deepfakes may ultimately avoid some of the pitfalls of other states, but the lack of precise definition in the bill itself may still find its way to court for enforcement.

The post Vermont attempts to regulate political AI deepfakes appeared first on Reason Foundation.

Source: https://reason.org/commentary/vermont-attempts-to-regulate-political-ai-deepfakes/