Mississippi AG tells Scott Colom to pull AI ad of Cindy Hyde-Smith under deepfake law

The attorney general gave a deadline

Mississippi Attorney General Lynn Fitch told Democratic U.S. Senate candidate Scott Colom on 23 October that an AI-generated campaign video of his opponent, Republican Sen. Cindy Hyde-Smith, may break the state's deepfake election law. According to the letter, quoted by the Mississippi Free Press, her office asked the campaign to remove the videos from every website and social account it controls by 3 p.m. that day, or the office may seek an injunction under section 97-13-47(4) of the Mississippi Code.

The letter went further than a takedown request. It warned that spreading the ads may violate section 97-13-47(2), which carries up to one year in prison, a fine of up to $5,000, or both. It told the campaign to preserve all related records and noted that destroying or withholding documents required by a grand jury subpoena is itself an offence. Fitch's office posted a short statement on deepfake political ads the same day.

Colom said he would comply, without conceding the legal point. In a statement reported by the Mississippi Free Press and the Magnolia Tribune, he called the video parody and said Hyde-Smith and her allies were trying to weaponize the justice system. He said he would not spend donors' money on costly litigation, adding: "We are going to delete the post in the spirit of being good stewards of their money."

What the video showed

Colom posted the ad on Facebook and Instagram on 1 October, 33 days before the 3 November election. It shows an AI-generated woman made to look like Hyde-Smith being chased by reporters with cameras and microphones, covering her face with a folder as papers fly, running through the halls and down the steps of the U.S. Capitol and sliding into a black car. A caption says the scene is based on a true story. Nothing in the ad says it was made with AI.

Colom has said the true story is a September encounter in which a reporter tried to ask Hyde-Smith about a grand jury's decision not to indict anyone over the death of Nolan Wells, and the senator declined and walked to a security guard. He told the Mississippi Free Press he was pointing out hypocrisy, because Hyde-Smith sits on a committee with jurisdiction over AI and data centres. In an earlier interview he said the campaign was trying to get young people's attention.

Neither side has been shy about synthetic content. Colom opened his campaign in September 2025 with a staged phone call to Hyde-Smith that was not labelled as fake, and the Republican Party asked federal prosecutors to look at that video. Hyde-Smith's campaign runs a site at colomforsenate.com, built to look like Colom's own, that attacks his record. The Mississippi Free Press noted that the site is not confirmed to use AI and also carries no disclaimer.

What Mississippi's 2024 law says

The complaint started with the Mississippi Republican Party. Chairman Mike Hurst sent Fitch a letter on 22 October asking her to investigate and seek an injunction, the Mississippi Free Press reported. Hurst argued that section 97-13-47, passed in 2024, makes it a crime to knowingly spread a deceptive AI-generated depiction of a candidate close to an election, and that a caption saying based on a true story is the opposite of a disclosure.

As described in the coverage, the law applies when a person knows the item is a digitization and the dissemination either happens within 90 days of an election, is done without the consent of the person depicted, or is done with intent to injure the candidate, influence the result or deter someone from voting. The basic penalty is up to a year in jail and a $5,000 fine. The Magnolia Tribune reported that Fitch's letter also cited a higher tier of up to five years and $10,000 when the intent is to deter people from voting.

The law also has a carve-out that sits at the centre of this fight. It does not cover content that constitutes satire or parody. Hurst's letter argued the video is not parody because it was made to look real and was not played for comic effect. Colom's campaign manager called it clearly a joke. Neither the attorney general's letter nor the campaign's response has been tested in court, and because Colom took the video down, a court may never decide which side is right.

Labels are the main tool states have

Mississippi is one of 31 states with laws requiring disclaimers on political ads that use AI-generated content, according to the Mississippi Free Press. The rules vary a lot. Pennsylvania recently went further than most and made advertisers label AI-generated images, video and audio in all ads, not just political ones. Mississippi's law focuses on deceptive depictions of candidates in the run-up to a vote, and a broader bill that would have let residents sue over unauthorised use of their name, likeness or voice died in committee in the 2026 session.

Scott Brennen, director of NYU's Center on Technology Policy, told the Mississippi Free Press earlier this month that most of these laws target deceptive deepfakes rather than what researchers now call slopaganda, obviously synthetic political content meant to mock or provoke rather than fool anyone. He said Colom's video could be read as the latter because most viewers would recognise it as fiction, and that labels are the front-line defence for deceptive content. He also said it would be up to the courts to decide whether either campaign's content breaks the law.

That is the uncomfortable gap this case exposes. A label requirement is easy to apply to a fake robocall or a doctored clip. It is much harder to apply to a cartoonish video that some viewers will see as a joke and others will take as footage, especially when the caption says it is based on real events. The parody exception was written to protect satire, but it gives any campaign caught without a label an obvious defence.

The precedent everyone cites

The case most often mentioned is the fake Joe Biden robocall that reached New Hampshire voters the night before the state's January 2024 primary, telling Democrats not to vote. The Federal Communications Commission later fined the political consultant behind it $6 million, and the telecom company that carried the calls agreed to a $1 million penalty. Brennen noted that the call got huge publicity but was identified as fake very quickly.

Since then the debate has moved from whether deepfakes will appear in campaigns to who is responsible for catching and labelling them. AI companies have added their own rules, such as Anthropic's election safeguards for Claude, and courts are separately wrestling with how to authenticate evidence that might be synthetic. Mississippi's case is a reminder that the most common enforcement tool right now is not a detection system but a letter from a state official.

What happens next

With the video gone, the injunction threat is likely moot, but the investigation the letter describes is not necessarily over. The attorney general's office has not said publicly whether it will pursue criminal charges, and the record-preservation demand suggests it wants to keep that option open. Any prosecution of a sitting district attorney running for the Senate would be politically explosive, and Colom's team has already framed the letter as an elected Republican using her office against a Democrat.

The election is on 3 November, with Hyde-Smith, Colom and independent candidate Ty Pinkins on the ballot. For other campaigns the practical lesson is narrower than the politics. In Mississippi, and in most of the states with similar laws, a short line saying a video was made with AI is cheap insurance, and leaving it off invites exactly this kind of letter in the final weeks of a race.

For voters, the lesson is to read captions sceptically. A video that says it is based on a true story can still be entirely synthetic, and the law in many states only requires a label if a regulator or court agrees the content was meant to deceive.