Skip to content
Article reader Listen + reading controls
LISTEN + READ YOUR WAY

Article reader

Preparing the reader…

0:00 0:00
Reading settings
Text size
100%

AI disclosures in political ads are necessary—and insufficient

A label on a political advertisement can tell us that artificial intelligence helped make it. It cannot tell us whether the message is true, who authorized the representation, how materially the content was altered, or whether millions of people saw it before the label appeared.

That is the problem the Artificial Intelligence (AI) Transparency in Elections Act tried to address in 2024. Senate Bill 3875 would have directed the Federal Election Commission (FEC) to require disclosures when covered political communications contained content “substantially generated” by AI. The bipartisan proposal recognized a real gap: voters could encounter a synthetic voice, image, or video without knowing that part of the apparent evidence had never occurred.

The bill advanced out of committee and reached the Senate calendar, but it did not become law before the 118th Congress ended. The later record makes the underlying design question more useful, not less. What would an effective disclosure regime need to accomplish—and what should no one expect a label to solve?

Revised and substantially expanded July 17, 2026, to incorporate the bill's legislative outcome and subsequent FEC action.

Disclosure answers one question

Political advertising already carries disclaimers identifying who paid for or authorized certain communications. An AI disclosure would add a different fact: whether software materially generated or altered what the audience sees or hears.

That information matters. A realistic recording of a candidate saying words the candidate never spoke makes a different claim on the viewer than an obviously satirical animation. Disclosure can interrupt that false inference. It can also give journalists, platforms, researchers, and election officials a common signal for investigating how synthetic media is being used.

But disclosure is not verification. A labeled fabrication remains a fabrication. An unlabeled authentic clip can still be edited out of context. A message can be generated without AI and remain intentionally deceptive. The mechanism should therefore be judged as one layer in an accountability system, not as a general solution to political misinformation.

The U.S. Government Accountability Office (GAO) reached a similar technical conclusion in its 2024 assessment of tools for combating deepfakes. Detection systems have limited effectiveness in real-world conditions, creators adapt to evade them, and identifying manipulated media does not guarantee that the resulting harm can be contained. A disclosure rule is useful precisely because technical detection alone is unreliable. It is inadequate for the same reason: it acts after a piece of media has already entered a social system optimized for rapid distribution.

“Substantially generated” is a difficult boundary

Most modern media passes through software. Campaigns remove background noise, correct color, translate text, generate captions, resize images, and test variations. Some of those uses improve accessibility or production quality without changing the substantive claim. Others manufacture a person's voice, appearance, conduct, or setting.

A workable rule has to distinguish between assistance and material representation. The important question is not whether AI touched the file. It is whether AI changed what a reasonable viewer would understand to have happened, who appeared to communicate, or what evidence the communication seemed to provide.

That standard still requires judgment, but it is closer to the public harm than a list of approved and prohibited tools. Models and editing workflows will change faster than legislation. The represented fact—who said or did what—changes much more slowly.

A sensible disclosure framework would therefore focus on materiality:

  • Did the content synthesize or materially alter a real person's voice, image, action, or statement?
  • Did it create a realistic event, place, document, or interaction that did not occur?
  • Would the alteration matter to a reasonable person's interpretation of the political claim?
  • Is the use clearly parody, illustration, translation, accessibility support, or routine production editing?

The disclosure should describe the material alteration in plain language. “AI-generated” is less useful than “This video contains a synthetic voice that does not record the candidate speaking.”

The FEC chose a narrower path

In September 2024, the FEC voted not to open an AI-specific rulemaking. Instead, it issued an interpretive rule explaining that the existing prohibition on fraudulent misrepresentation is technology-neutral and can apply when AI-assisted media is used to commit the conduct already covered by law.

That clarification closed one ambiguity but left a broader gap. The underlying statute addresses particular forms of fraudulent misrepresentation involving candidates, political parties, and solicitation. It is not a general provenance or disclosure requirement for synthetic political media.

This is a recurring problem in technology governance. Existing law often applies to harmful conduct regardless of the tool. That is valuable; a new technology should not create an exemption from an old responsibility. But technology-neutral enforcement does not always supply the new information infrastructure that a changed media environment requires.

The public needs both: enforcement against prohibited conduct and reliable information about the provenance of consequential media.

A label needs an accountability chain behind it

The strongest disclosure regime would not end with pixels placed on an advertisement. It would connect the visible notice to records capable of supporting review and enforcement.

That chain should include:

  1. Identity and authorization. Who paid for the communication, who approved it, and which entity is responsible for the disclosure?
  2. Material-use description. What was generated or altered, and which human representation or factual event was affected?
  3. Durable placement. Does the disclosure remain visible when a clip is shortened, reposted, embedded, or viewed without sound?
  4. Machine-readable provenance. Can platforms and investigators preserve and transmit the disclosure as the media moves between systems?
  5. Public availability. Is there an accessible archive of covered advertisements, sponsors, distribution periods, material alterations, and corrections?
  6. Correction and enforcement. Who can require a corrected disclosure, how quickly must it appear, and what happens when a sponsor knowingly omits or falsifies one?

Technical provenance standards can help, but they are not self-enforcing. Metadata can be stripped. Authenticity credentials can show that a file came from a particular source without showing that its political claim is accurate. The accountable object is therefore not only the media file; it is the chain connecting sponsor, production decision, distribution, disclosure, and remedy.

Measure whether disclosure helps voters

Passing a rule and counting labeled ads would show administrative activity. It would not show whether the intervention improved public understanding.

A serious evaluation would test whether people notice the disclosure, understand what it means, distinguish material synthesis from routine editing, and retain the correction after exposure. It would also examine whether a standard label creates a “liar's dividend”—a tendency to dismiss authentic evidence merely because synthetic media exists.

The results should shape presentation. A notice placed after a video, hidden behind a platform menu, or expressed in legal language may satisfy a formal requirement while arriving too late to affect belief. The design problem is behavioral as well as statutory.

Build for provenance, not panic

The 2024 bill was right to identify undisclosed synthetic political media as a transparency problem. It was also only a starting point.

AI should not become a vague warning attached to every digitally edited communication. Nor should campaigns be permitted to create realistic false evidence while treating the production method as irrelevant. The durable middle ground is a materiality standard, a clear explanation of what changed, and a verifiable accountability chain behind the visible label.

Election trust cannot be restored by a badge in the corner of a screen. It can be strengthened when voters can identify the speaker, understand how the apparent evidence was made, trace who authorized it, and see meaningful consequences when those representations are deliberately false.

Sources

READER-NEUTRAL SUBSCRIPTION

Follow Field Notes via RSS.

Copy this address into the RSS reader you already use. New notes will appear there automatically—no account, email address, or tracking required.

Open the feed