AI Political Content and Deepfake Regulation: Rules for Synthetic Politics: Deepfake laws, disclosure duties, robocall bans

In brief

AI and deepfake rules are separate instruments, and each one binds a different party: the company that builds the system, the organisation that uses it, the committee that buys the ad, the caller who dials. That choice of addressee, not the rule's severity, decides who is exempt and what survives. Europe's disclosure duty exempts private individuals, while America's robocall rule turns on whether the line is a landline or a mobile. Four US cases so far show duties aimed at the general public struck down and one binding a regulated committee untouched, though no court has named that pattern as a doctrine.

How to use this

Before treating a new AI speech rule as binding, identify the party it names: the provider that builds the model, the deployer that uses it, the buyer of the advertisement, or the caller who dials, and check whether you fall inside or outside that description. If you are a private individual posting synthetic content for yourself, check the personal-activity exemption before assuming a European disclosure duty applies. If you are calling voters, count the calls per landline and take consent before dialling a mobile. When drafting, aim duties at regulated political spenders rather than the general public, and do not rely on an AI-specific duty being enforced: check the record first.

What the story is about

Two days before the 2024 New Hampshire Democratic presidential primary, voters' phones rang with a cloned voice of President Biden. The Federal Communications Commission fined the man behind the calls: "We impose a penalty of $6,000,000 against Steve Kramer" (Federal Communications Commission, 2024d). The voice is not what he was charged for. The order finds him "in violation of the Truth in Caller ID Act of 2009", because the calls carried the telephone number of a New Hampshire political operative who had neither known about them nor agreed to them. The most discussed political deepfake in the world was punished as a telephone-numbering offence. No collection action on the penalty appears in the federal docket record.

Rules that do name AI exist, and each attaches its duty to a different party. Europe's AI Act splits its deepfake rule between two parties. Article 50(2) binds the provider, meaning the company that builds a system generating audio, images, video or text: its output must be "marked in a machine-readable format and detectable as artificially generated or manipulated" (European Union, 2024b). The rule assumes such a mark can be traced later; whether it can is the subject of our article about detecting deepfakes. Article 50(4) binds the deployer, meaning the organisation that uses the system: where the output is a deep fake, it "shall disclose that the content has been artificially generated or manipulated". The Act then exempts the private individual: "This Regulation does not apply to obligations of deployers who are natural persons using AI systems in the course of a purely personal non-professional activity" (European Union, 2024b). So a person who makes and posts an election deepfake purely for themselves owes no disclosure. A party, a campaign or a paid staffer does.

These European duties have only just begun. When the EU simplified the AI Act in July 2026, it left Article 50 on its original start date of 2 August 2026 (European Union, 2026). What it delayed were the heavier duties on high-risk systems, a class that includes "AI systems intended to be used for influencing the outcome of an election or referendum or the voting behaviour of natural persons in the exercise of their vote in elections or referenda" (European Union, 2024b). Those now apply from 2 December 2027, sixteen months later than first planned. A breach of Article 50 can cost up to EUR 15,000,000 or 3% of total worldwide annual turnover, whichever is higher. So the deepfake duty is running now; the election-specific duties are fifteen months away.

Europe's political-advertising regulation is a separate law with a different addressee. Regulation (EU) 2024/900 binds the buyer of a political advertisement, and its operative text mentions AI in one place only: the ad's notice must say "whether an artificial intelligence system has been used to target or deliver the political advertisement" (European Union, 2024a). Nothing in that regulation addresses AI-generated content. Our article about the EU's political advertising rules covers the targeting and delivery side.

America's robocall rule binds the caller, and it comes from an old law read in a new way. In February 2024 the Federal Communications Commission confirmed that the restrictions on artificial or prerecorded voices in the Telephone Consumer Protection Act of 1991 "encompass current AI technologies that generate human voices", so such calls need the called person's prior express consent (Federal Communications Commission, 2024a). The duty then turns on which line rings. A call to a home landline needs no consent at all if it "Is not made for a commercial purpose and the caller makes no more than three calls within any consecutive 30-day period to the residential line" and the caller honours requests to stop (Office of the Federal Register, 2026). Political calls are not commercial. A call to a mobile needs prior express consent. So a campaign may place three AI-voice calls a month to a landline without asking, provided it honours opt-outs, and none to a mobile.

Two federal proposals would have gone further, and neither became a rule. The FCC proposed making callers disclose when a call is AI-generated (Federal Communications Commission, 2024b). It separately proposed an on-air announcement for political ads containing AI-generated content, binding broadcast stations and cable operators, so not the internet (Federal Communications Commission, 2024c). Comments on both closed in 2024, and no final rule has issued. The Federal Election Commission turned down a petition for an AI rule (Federal Election Commission, 2024a) and issued an interpretive rule instead, holding that "The statute, and the Commission's implementing regulation, is technology neutral". That rule says of itself: "It does not bind the Commission or any members of the general public, nor does it create or remove any rights, duties, or obligations" (Federal Election Commission, 2024b).

In the courts, the provisions that reach the general public have been falling. California's AB 2839 covered materially deceptive election content, let anyone who received it sue, and required disclaimers even on satire and parody. A federal court held that "AB 2839 fails strict scrutiny in its entirety" (Kohls v. Bonta, 2025b), strict scrutiny being the most demanding test a law restricting speech can face. A second California law, AB 2655, which required large platforms to remove deceptive election content, fell "because AB 2655 violates and is preempted by Section 230 of the Communications Decency Act of 1996" (Kohls v. Bonta, 2025a), the federal law that shields online platforms. Hawaii's law went too: "this Court concludes that Act 191 is unconstitutionally vague pursuant to the Fourteenth Amendment" (Babylon Bee, LLC v. Lopez, 2026). California's third AI duty is untouched. Government Code § 84514 requires a committee, a regulated political spender that already carries disclosure duties, to put the words "Ad generated or substantially altered using artificial intelligence" on its qualifying ads, and it has never been challenged (California, 2024). The duties reaching anyone who distributes a fake were struck down, and the one binding a committee stands. That is a pattern in four cases, not a doctrine; no court has said it.

So each rule names its own party. Europe's marking duty binds the company that builds the AI, and its disclosure duty the organisation that uses it. The advertising duties bind whoever buys the ad. The American call rules bind the caller who dials. In the cases decided so far, duties aimed at organised political spenders have survived in court, and those reaching the general public were struck down. Whom a rule binds matters more than how tough it is.

So what

For the people these rules land on, the first question about any new one is whom it binds. What follows differs for a party paying for campaign material and for a government writing the rules.

For political parties

If you pay for political ads in the United States, the state layer is where most duties sit. "Thirty-one states have enacted laws regulating the use of deepfakes in political messaging", by the count of the National Conference of State Legislatures, a tracker that publishes its counting rule (National Conference of State Legislatures, 2026). Twenty-eight of them require a disclosure on the media itself, modelled on the paid-for disclaimers ads already carry. Minnesota and Texas instead ban publication within a window before an election, and Maryland bans it all year. Colorado and Utah also require the disclosure in the file's metadata, the hidden data that travels with an image or video.

Does carrying a label achieve anything? Whether a label changes what voters believe is a separate question, which our article about AI disclosure labels covers; this one reports what the duties require. What a disclosure duty itself achieves has been argued rather than measured. Fisher (2025), writing in a peer-reviewed philosophy journal, makes the case for labelling on grounds that do not depend on changing anyone's mind: people have an interest in knowing which content is synthetic, and "labelling synthetic content in public forums is an essential first step". Block (2026), analysing the AI Act's neighbouring duty for systems that interact with people, warns that transparency "risks becoming a symbolic rather than a functional safeguard under the current regulatory design". Neither measures anything. The evidence contains no measurement of compliance with these duties, or of their effect.

The only look at labelling in a real European campaign comes from Germany. Geise et al. (2025) analysed Instagram posts from the main German parties and their youth wings in the six weeks before the February 2025 federal election. The far-right AfD used synthetic images significantly more than other parties, and the images were "predominantly photorealistic and often lack clear labeling" (Geise et al., 2025). That campaign ran eighteen months before the EU's disclosure duty applied, so the study shows what parties did voluntarily, not whether they obey a law. It is a baseline to measure the duty against, not a compliance rate.

For government

The record holds no enforcement decision under any AI-specific rule described here: none under Article 50, which had applied for six weeks, and none under any of the thirty-one state statutes. In Europe, the penalties for breaching Article 50 are set by each member state, and no consolidated list of those national penalties exists (European Union, 2024b). So far, each of these duties is a rule on paper.

Three questions remain open. The first is whether the Ninth Circuit, which is hearing the appeal over California's law, agrees with the district court (Kohls v. Bonta, 2025b). The second is whether a disclosure duty aimed at a regulated committee really does survive where one aimed at the public does not: § 84514 has never been challenged, so the distinction has never been tested (California, 2024). The third is whether the sixteen-month delay to the election high-risk duties was deliberate; the amending regulation gives its reasons for the delay, and elections are not among them (European Union, 2026). On this record, binding a regulated spender rather than the general public looks the safer course for a drafter, though no court has tested it.

Case studies

Christopher Kohls challenged both California's law and Minnesota's, and got opposite results. In California the statute was struck down in its entirety on the merits (Kohls v. Bonta, 2025b). Minnesota's law makes it a crime to spread a deep fake within 90 days before an election with intent to injure a candidate or influence the result; a state or local candidate who breaks it forfeits the nomination or office (Kohls v. Ellison, 2026). There the courts never reached free speech. Kohls was held to lack standing because his videos fell outside the statute, and his co-plaintiff to have waited more than sixteen months. The Eighth Circuit affirmed, noting that "The district court did not rule on the constitutionality of the statute" (Kohls v. Ellison, 2026). Minnesota's is the harsher law, and it is still in force. Whether it binds you depended partly on who sued, where, and how quickly.

References

Babylon Bee, LLC v. Lopez (2026) United States District Court for the District of Hawaii, Civ. No. 1:25-cv-00234-SASP-KJM, order granting plaintiffs' motion for summary judgment, filed 30 January. Available at: https://storage.courtlistener.com/recap/gov.uscourts.hid.174048/gov.uscourts.hid.174048.47.0.pdf (Accessed: 14 September 2026).

Block, M.J. (2026) 'Transparency in human-AI interaction: an analysis of Article 50(1) AI Act', Computer Law & Security Review, 61, 106327. Available at: https://doi.org/10.1016/j.clsr.2026.106327 (Accessed: 14 September 2026).

California (2024) Government Code § 84514: disclosure in advertisements. Available at: https://leginfo.legislature.ca.gov/faces/codes_displaySection.xhtml?lawCode=GOV&sectionNum=84514 (Accessed: 14 September 2026).

European Union (2024a) Regulation (EU) 2024/900 of the European Parliament and of the Council of 13 March 2024 on the transparency and targeting of political advertising, OJ L, 2024/900, 20.3.2024, Article 19. Available at: http://data.europa.eu/eli/reg/2024/900/oj (Accessed: 14 September 2026).

European Union (2024b) Regulation (EU) 2024/1689 of the European Parliament and of the Council of 13 June 2024 laying down harmonised rules on artificial intelligence (Artificial Intelligence Act), OJ L, 2024/1689, 12.7.2024, Articles 2, 3, 50, 70, 99 and 113 and Annex III. Available at: http://data.europa.eu/eli/reg/2024/1689/oj (Accessed: 14 September 2026).

European Union (2026) Regulation (EU) 2026/1744 of the European Parliament and of the Council of 8 July 2026 amending Regulations (EU) 2024/1689, (EU) 2018/1139 and (EU) 2023/1230 as regards the simplification of the implementation of harmonised rules on artificial intelligence (Digital Omnibus on AI), OJ L, 2026/1744, 24.7.2026. Available at: http://data.europa.eu/eli/reg/2026/1744/oj (Accessed: 14 September 2026).

Federal Communications Commission (2024a) In the matter of implications of artificial intelligence technologies on protecting consumers from unwanted robocalls and robotexts: declaratory ruling, FCC 24-17, CG Docket No. 23-362, adopted 2 February 2024, released 8 February 2024. Available at: https://docs.fcc.gov/public/attachments/FCC-24-17A1.pdf (Accessed: 14 September 2026).

Federal Communications Commission (2024b) In the matter of implications of artificial intelligence technologies on protecting consumers from unwanted robocalls and robotexts: notice of proposed rulemaking and notice of inquiry, FCC 24-84, CG Docket No. 23-362, adopted 7 August 2024, released 8 August 2024, published at 89 FR 73321, 10 September 2024. Available at: https://www.federalregister.gov/documents/2024/09/10/2024-19028/implications-of-artificial-intelligence-technologies-on-protecting-consumers-from-unwanted-robocalls (Accessed: 14 September 2026).

Federal Communications Commission (2024c) Disclosure and transparency of artificial intelligence-generated content in political advertisements: notice of proposed rulemaking, FCC 24-74, MB Docket No. 24-211, 89 FR 63381, 5 August 2024. Available at: https://www.federalregister.gov/documents/2024/08/05/2024-16977/disclosure-and-transparency-of-artificial-intelligence-generated-content-in-political-advertisements (Accessed: 14 September 2026).

Federal Communications Commission (2024d) In the matter of Steve Kramer: forfeiture order, FCC 24-104, File No. EB-TCD-24-00036094, adopted 26 September 2024, released 30 September 2024. Available at: https://docs.fcc.gov/public/attachments/FCC-24-104A1.pdf (Accessed: 14 September 2026).

Federal Election Commission (2024a) Artificial intelligence in campaign ads: notification of disposition of petition for rulemaking, Notice 2024-23, 89 FR 78826, 26 September 2024. Available at: https://www.federalregister.gov/documents/2024/09/26/2024-21979/artificial-intelligence-in-campaign-ads (Accessed: 14 September 2026).

Federal Election Commission (2024b) Fraudulent misrepresentation of campaign authority: interpretive rule, Notice 2024-24, 89 FR 78785, 26 September 2024, effective 26 September 2024. Available at: https://www.federalregister.gov/documents/2024/09/26/2024-21983/fraudulent-misrepresentation-of-campaign-authority (Accessed: 14 September 2026).

Fisher, S.A. (2025) 'Something AI should tell you: the case for labelling synthetic content', Journal of Applied Philosophy, 42(1), pp. 272–286. Available at: https://doi.org/10.1111/japp.12758 (Accessed: 14 September 2026).

Geise, S., Luther, A.R., Reich, S. and Linke, M. (2025) 'A new face of political advertising? Synthetic imagery in the 2025 German federal election campaigns on social media', Studies in Communication and Media, 14(4), pp. 485–516. Available at: https://doi.org/10.5771/2192-4007-2025-4-485 (Accessed: 14 September 2026).

Kohls v. Bonta (2025a) United States District Court for the Eastern District of California, No. 2:24-cv-02527-JAM-CKD, order and final judgment and permanent injunction as to AB 2655, filed 20 August. Available at: https://storage.courtlistener.com/recap/gov.uscourts.mnd.224575/gov.uscourts.mnd.224575.42.1.pdf (Accessed: 14 September 2026).

Kohls v. Bonta (2025b) United States District Court for the Eastern District of California, No. 2:24-cv-02527-JAM-CKD, order granting plaintiffs' motion for summary judgment (AB 2839), filed 29 August. Available at: https://storage.courtlistener.com/recap/gov.uscourts.caed.453046/gov.uscourts.caed.453046.101.0.pdf (Accessed: 14 September 2026).

Kohls v. Ellison (2026) United States Court of Appeals for the Eighth Circuit, No. 25-1300, submitted 22 October 2025, filed 9 February. Available at: https://storage.courtlistener.com/recap/gov.uscourts.mnd.220348/gov.uscourts.mnd.220348.66.0.pdf (Accessed: 14 September 2026).

National Conference of State Legislatures (2026) Deepfakes in elections and campaigns, updated 23 June. Available at: https://www.ncsl.org/elections-and-campaigns/deepfakes-in-elections-and-campaigns (Accessed: 14 September 2026).

Office of the Federal Register (2026) 47 CFR § 64.1200: delivery restrictions. Available at: https://www.ecfr.gov/current/title-47/chapter-I/subchapter-B/part-64/subpart-L/section-64.1200 (Accessed: 14 September 2026).