The case for stronger rules is real
Al Jazeera reported on 9 October 2026 that candidates and their backers have spent more than $80 million on more than 164 AI-based ads this election season, citing the Wesleyan Media Project. Republican candidates and their backers were behind about 80 percent of them. The same report notes that there is no federal law regulating AI deepfake ads, while 31 states require AI-made ads to be labelled.
The strongest argument for tougher regulation starts there. Voters decide in short windows, and a convincing fake of a candidate saying something they never said can circulate before any correction arrives. A patchwork of state laws is hard for national campaigns to follow and easy for out-of-state actors to ignore. A federal statute would at least give everyone one rulebook. That argument deserves a serious answer, not a dismissal.
The answer is that the evidence in the hook points toward a narrower, more durable approach than blanket bans.
What federal law already does
The Congressional Research Service put the gap plainly in a September 2024 Insight: "No federal statute or regulation specifically addresses artificial intelligence (AI) in political campaigns." Congress has not changed that. CRS noted that a Senate Rules Committee bill, S. 3875, which would have required AI disclaimers on certain political ads, was reported in May 2024.
Two agencies have acted within their existing authority.
- The FEC. On 19 September 2024 the Commission voted not to open a rulemaking on AI in campaign ads. It adopted an interpretive rule instead, saying that 52 U.S.C. § 30124 and 11 CFR 110.16 "apply irrespective to the technology being used to conduct fraudulent misrepresentation." In the Commission's words, the statute "is technology neutral." It will apply the law to specific technologies case by case.
- The FCC. On 8 February 2024 the FCC announced a unanimously adopted Declaratory Ruling that calls made with AI-generated voices are "artificial" under the Telephone Consumer Protection Act. The ruling took effect immediately. It was a response to AI voice-cloning robocalls, and it works because it extends a long-standing consent rule instead of creating a new speech category.
Both actions share a design principle. They target conduct that was already unlawful, such as fraudulent misrepresentation of who is speaking or unconsented automated calls. They do not target a technology or a viewpoint.
Why labelling is a weak tool
Labelling laws are the most popular state answer, and they are modest and defensible. But the one enforcement step Al Jazeera describes shows their limits. The New York Democratic Party has filed a complaint with the state board of elections saying Bruce Blakeman's AI ads violate disclosure requirements and are "misleading to the voters." The ads showed Zohran Mamdani and Kathy Hochul "gardening together and collaborating to raise prices for residents," though neither appeared in them. Blakeman defended the ads, saying they were labelled as AI-generated, with a disclaimer "for a few seconds at the start," and that they follow "a long tradition of satire and parody."
That dispute is a fair preview of every future one. It turns on how long a disclaimer must stay on screen, how prominent it must be, and where satire ends and deception begins. Those are fact-heavy questions that a board of elections will answer after the ad has run, and possibly after the election. Al Jazeera also quotes NYU's Scott Babwah Brennen describing obviously fake ads, which he calls "slopaganda," as the defining feature of this cycle. A rule that requires a label does little against content nobody mistakes for real.
The partisan split deserves care too. With about 80 percent of the counted ads tied to Republicans, any enforcement regime will be accused of partisan targeting. That is a reason to prefer rules a neutral clerk can apply mechanically, such as a clear disclosure standard or the FEC's fraud provision, over discretionary judgments about whether an ad is "deceptive."
What happens when states go further
California tried the broader approach. Its AB 2839 restricted materially deceptive election content, and a federal court enjoined it on 2 October 2024. According to the Tech Policy Press litigation tracker, the court wrote that the law "does not use the least restrictive means available for advancing the State's interest here" and that "counter speech is a less restrictive alternative to prohibiting videos such as those posted by Plaintiff, no matter how offensive or inappropriate someone may find them." The tracker also records that on 3 January 2025 the court stayed enforcement of AB 2655, a companion law aimed at platforms, in its entirety. We have not verified the later procedural history of that case, so we treat the 2024 orders as the reliable record.
The lesson is not that states can do nothing. It is that laws which ask officials or platforms to decide what is true will keep meeting First Amendment limits, and platform-facing mandates add intermediary-liability problems on top. Pushing platforms to remove political speech on government say-so would be worse for the open internet than the deepfakes themselves.
A proportionate agenda
- Enforce existing fraud law first. The FEC has said § 30124 reaches AI-assisted fraudulent misrepresentation. A test case against an ad that falsely purports to come from a campaign would show whether that guidance has teeth.
- Keep disclosure rules narrow and objective. Specify a minimum on-screen duration and size for AI disclosures, and exempt obvious satire and parody. This makes complaints like the New York one resolvable quickly.
- Prefer speed and counter-speech to takedowns. Rapid-response correction by the targeted candidate, with platforms offering context labels voluntarily, works within First Amendment doctrine as California's litigation shows.
- Legislate federally only for the clear cases. Cloned-voice and fabricated-endorsement impersonation are fraud. A federal bill limited to them would end the patchwork without licensing viewpoint policing.
The midterm numbers show how fast the technology moved into campaigns. The response that will last is the one built on fraud, objective disclosure and counter-speech, because it is the one courts have left standing.