Political deepfakes: documented cases and what the election records show

This page sits in the deepfake detection hub and applies the rule of the documented deepfake incidents page to elections: a case appears only if a regulator, court or election authority has described it in a document anyone can read. Press accounts, campaign claims and vendor reports are not sources here.

01Why only official records?

Election seasons produce more claims of deepfakes than deepfakes with evidence. A clip can be called fake by one side and authentic by the other, and a detector score settles nothing, because the published in-the-wild results put the best off-the-shelf open-source video detector at 0.63 AUC on fakes that circulated in 2024 (see the checking a face video or photo yourself page for the order of evidence). An official record is different. A regulator's order states what was sent, when and to whom, after an investigation with legal consequences for getting it wrong. A court order states what a statute says and what the court found. Those are the only documents this page cites.

That rule has a cost. It leaves out widely reported incidents in several countries for which no regulator or court has published a description, and it means this page is short. A short list with sources is more useful to a reader checking a claim than a long one without them.

02The New Hampshire robocalls, January 2024

The one political deepfake with a complete regulator file is the New Hampshire robocall. The FCC Enforcement Bureau's order of 21 August 2024 states that on 21 January 2024, "two days before the New Hampshire Democratic primary election", potential voters "received thousands of spoofed robocalls carrying a deepfake generative artificial intelligence (AI) voice message purporting to be from the president of the United States", and that "the prerecorded AI message, using voice cloning technology, told recipients not to vote in the upcoming Primary Election" (FCC DA 24-790).

The enforcement target in that order was the carrier, not the person who commissioned the calls. Lingo Telecom, LLC had signed the calls with caller ID attestations under 47 CFR 64.6301(a); a Notice of Apparent Liability of 23 May 2024 proposed a US$2,000,000 penalty, and the consent decree adopted on 21 August 2024 set a US$1,000,000 civil penalty and a compliance plan. A separate forfeiture proceeding against the individual who commissioned the calls exists but names a private person and is not cited here.

Two weeks after the calls, the Commission closed the legal gap the case exposed. Its Declaratory Ruling FCC 24-17, adopted 2 February and released 8 February 2024, confirmed "that the TCPA's restrictions on the use of 'artificial or prerecorded voice' encompass current AI technologies that generate human voices", so that a cloned-voice call needs the prior express consent of the called party, and noted that "no commenter opposes these conclusions" (FCC 24-17). The ruling's reasoning is plain: cloned voices "are 'artificial' voice messages because a person is not speaking them".

03A court record: the California parody case, 2024

The second record is a court order rather than an enforcement action, and it documents a political deepfake from the other direction: a state law written against them, and a video that tested it. On 2 October 2024 the United States District Court for the Eastern District of California granted a preliminary injunction against California's AB 2839, "Elections: deceptive media in advertisements", which had been signed on 17 September 2024 (order of 2 October 2024, case 2:24-cv-02527, Document 14).

The order describes the statute. AB 2839 barred the knowing, malicious distribution of "materially deceptive content" portraying a candidate "as doing or saying something that the candidate did not do or say", for "the 120 days before any election in California and, except for depictions of a candidate, for 60 days after the election". Parody and satire were exempt only if labelled with a disclosure "no smaller than the largest font size of other text appearing in the visual media". The suit was brought by a content creator over a parody video depicting Vice President Kamala Harris posted on 26 July 2024, which the complaint says drew over 100 million views. The court held that the law "does not pass constitutional scrutiny because the law does not use the least restrictive means available", and enjoined its enforcement "except for the audio only severed portion of the statute".

The order is the primary record of three facts: the text of the statute, its dates, and the court's finding. It does not decide whether the video was a deepfake in the technical sense; the state's own definition in the record is "a manipulated piece of media where a person's likeness, image or void is digitally created or swapped with another person's". The deepfake laws page tracks how other statutes draw that line.

04What have election authorities said?

The European Commission's guidelines for very large online platforms on electoral risks, published in the Official Journal on 26 April 2024 under Article 35(3) of the Digital Services Act, are the clearest election-authority document on deepfakes that could be read for this page. They ask platforms to "clearly label, or otherwise make distinguishable through prominent markings, synthetic or manipulated images, audio or videos that appreciably resemble existing persons", to provide tools for users to label AI content, to "make sure the labelled generative AI content retains its label once it is shared by other users on the platform", and to be able to react rapidly to "manipulated images, voice recordings or deepfakes, for example of political actors contending in elections" (Commission guidelines C/2024/3014). What a retained label or a found mark proves is set out on the AI video watermarks page. The guidelines point forward to the AI Act's marking duties, which the EU AI Act's Article 50 labelling duties page sets out.

One more official record touches political figures without an election. The Hong Kong Government's Legislative Council reply of 26 June 2024 states that between 1 November 2023 and 31 May 2024 "a total of 21 online deepfake video clips involving impersonation of government officials or celebrities were identified by or reported to the Police", all removed at the Police's request, and that no member of the public was reported defrauded by them (LCQ9, 26 June 2024).

05What do the political deepfake records have in common?

Three patterns stand out. The only enforced case was audio, not face video, and it was caught through telephone rules on caller ID and consent rather than by any detector. The court record concerns a law about deepfakes rather than a finding that a particular video was one. And none of the five records relies on a detection score; each rests on who sent what, when, and on the text of a rule. That matches the lesson of this site's face dataset consent history: the facts that hold up are the ones with a document behind them.