AI Content and Idaho's Synthetic Media Law

A two-person campaign can now produce what used to need a design shop, a voice actor and an editor. It can also produce, in an afternoon, a convincing recording of an opponent saying something they never said. Idaho's Legislature, the FEC, the FCC and every major platform have each written rules for the second thing while leaving the first mostly alone. This lesson goes through those rules and then turns them into house rules a small campaign can follow without a compliance staff. The technical side of how AI text and images are made and detected is in the AI detection module's lesson on watermarks and provenance; the robocall side is in lesson 7; and the ordinary "paid for by" rules that apply to every ad, AI or not, are in lesson 6.

Where AI shows up in small campaigns

Most campaign use of AI is unremarkable and raises no legal question at all:

• Text: first drafts of a fundraising email, a summary of a 90-page budget, ten versions of a headline to test. A person still checks every fact; AI writing tools invent figures, quotes and citations with complete confidence.
• Translation: a Spanish version of a mailer or website page. Have a fluent speaker read it before it goes out.
• Images: an illustration for a post about a road bond, a background, a stylized graphic. Jeff's ImageStudio makes AI images at 1024x1024, 1080x1920 and 1920x1080, which are square, vertical and horizontal social sizes.
• Voice: text to speech for a narrated explainer, as in Jeff's Voice Studio, or a candidate cloning their own voice to make versions in another language.
• Video: cleaning up audio, removing background noise, generating captions, and less often, generated footage.

The legal line runs through realism and identity. A cartoon of a pothole is fine. A realistic video of a named county commissioner standing in that pothole, saying words he never said, is the thing every rule below is aimed at.

Idaho's FAIR Elections Act, Idaho Code 67-6628A

How it passed

The bill began in January 2024 as House Bill 407 and then House Bill 426, brought by House Minority Leader Ilana Rubel and Rep. Bruce Skaug, a bipartisan pair. The Idaho Capital Sun reported on January 25, 2024 that the House State Affairs Committee sent the revised version to the floor without a recommendation; some members wanted a set money penalty and others worried it did not reach non-candidates. The version that became law was House Bill 664, sponsored by the State Affairs Committee. The House passed it 61 to 8 on March 7, 2024, the Senate 23 to 11 on March 18, and Gov. Brad Little signed it March 25, 2024, effective that day, as Session Laws chapter 172. The section's history line shows no amendment in 2025 or 2026.

What it says

Section 67-6628A sits in the Sunshine Law chapter and works in pieces:

• Synthetic media (subsection 2(c)) is "an audio recording or a video recording of an individual's speech or conduct" made with generative adversarial network techniques "or other digital technology" to create realistic but false audio or video. It must appear to a reasonable person to show real speech or conduct that did not happen, and, in a paraphrase of the statute's stronger wording, give a basically different impression than the unaltered original would.
• Injunction (subsection 3): a candidate whose action or speech is deceptively represented through synthetic media in an electioneering communication may ask a court to stop its publication.
• Damages (subsection 4): the candidate may sue the information content provider for general or special damages, and the court may award reasonable fees and costs to the prevailing party. Other remedies are not limited.
• The disclosure defense (subsection 5): it is an affirmative defense that the communication included the statement "This (video/audio) has been manipulated." For video, the text must be prominently displayed in a size the average viewer can easily read, for the full duration. For audio-only media, it must be clearly spoken at an easily heard pitch at the beginning and end, and repeated at intervals of no more than two minutes if the audio runs longer than two minutes.
• Burden (subsection 6): the candidate must prove the use of synthetic media by clear and convincing evidence, a higher bar than the usual civil standard. Courts are told to move quickly (subsection 7).
• Who is liable (subsections 8 to 10): the information content provider, not the medium that carries the message, unless the medium removes the disclosure or changes the content so it becomes synthetic media. Federally licensed broadcasters transmitting ads under 47 U.S.C. 315 are excluded from that exception, and a provider or user of an interactive computer service is not treated as the publisher of content someone else provided.

The electioneering link and the window

The statute has no 90-day window of its own. It reaches synthetic media "in an electioneering communication," and that term is defined in Idaho Code 67-6602(6): a communication that unambiguously refers to a candidate, is broadcast, printed, mailed, delivered or otherwise distributed within 30 days before a primary or 60 days before a general election, and reaches the voters for that office. So in practice the Idaho remedy covers fakes about candidates in the last month before a primary and the last two months before a general.

The definition also excludes several things in 67-6602(6)(b), including news stories and editorials in outlets not owned by a candidate, committee or party, communications to an organization's own members, and communications that are an "expenditure" or "independent expenditure" under the chapter. Read literally, that last exclusion raises a question the statute does not answer: whether a deceptive deepfake paid for as a campaign expenditure or an independent expenditure is outside the FAIR Act's reach, even though it is the kind of ad the law seems aimed at. We found no Idaho court decision or Secretary of State guidance resolving it as of October 2026. A campaign should not plan around a reading that a court has never adopted; the safe practice is to treat any realistic fake of a candidate as covered.

Situation	Inside 67-6628A's text?	What else applies
Realistic AI video of a candidate saying words they never said, posted 20 days before the primary	Yes, if it reaches that office's voters	Platform takedown rules; defamation law; the disclosure is a defense
Same video posted 120 days before the general	No, outside the 30/60-day window	Defamation; platform rules; the political cost of being caught
AI still photo of a candidate at a protest that never happened	No, the definition covers audio and video recordings	Defamation and privacy claims; platform rules on realistic altered images
Cloned voice of a ballot measure opponent who is not a candidate	No, only candidates may sue	Defamation; FCC TCPA rules if used in calls (lesson 7)
Candidate uses AI to translate her own video into Spanish in her own cloned voice	No deceptive representation of a candidate	Platform disclosure rules for realistic altered content; good practice says label it
Obvious cartoon of a candidate as a superhero	No, not realistic	Satire and parody are protected speech and exempt under most platform rules
AI-drafted text of a mailer	No, text is not audio or video	"Paid for by" rules (lesson 6); accuracy is on you

Federal rules

The FEC

Federal law has long barred a federal candidate or agent from fraudulently pretending to speak for another candidate or party in a damaging way, and bars anyone from pretending to speak for a candidate to raise money: 52 U.S.C. 30124, with the regulation at 11 CFR 110.16. In 2023 an advocacy group petitioned the FEC to write AI-specific rules. On September 19, 2024, the Commission declined to open a rulemaking and instead adopted an interpretive rule saying the statute and regulation are "technology neutral": fraud done with AI is covered the same as fraud done any other way, decided case by case. The Congressional Research Service noted that the actions did not require any new disclaimers on political ads. All of this applies to federal races, such as Idaho's congressional and U.S. Senate seats; state and local races rely on Idaho law.

The FCC

On February 8, 2024, the FCC announced a unanimous ruling (Declaratory Ruling FCC 24-17, adopted February 2) that calls using AI-generated voices are "artificial" under the Telephone Consumer Protection Act, effective immediately. An AI-voiced robocall needs the same consent as any other prerecorded or artificial voice call, and state attorneys general can enforce it. The rules for campaigns, the New Hampshire primary case that prompted much of the attention, and the consent requirements are in lesson 7.

In July 2024 the FCC voted 3 to 2 to propose rules that would have broadcasters ask political advertisers whether ads contain AI-generated content, announce it on air and note it in their political files. News reports at the time noted that the FCC lacks authority over online and streaming ads, so the proposal would not have reached them. As of October 2026, we found no final rule adopted; treat it as a proposal, not a requirement.

Platform rules as of October 2026

The platforms' rules are stricter than the law in most places, and they apply to every race, federal, state and local, ballot measure or candidate. They also change without notice. Each line below comes from the platform's page or a dated report fetched in October 2026.

Platform	What must be disclosed	How	What is exempt or barred
Meta (Facebook, Instagram) ads about social issues, elections or politics	Photorealistic image or video, or realistic audio, digitally created or altered, in effect since January 2024	A checkbox in ad setup; Meta adds a label. Failing to disclose risks ad removal or bans (Social Media Today)	Meta's own help page could not be loaded for this lesson; check the current exemptions in Ads Manager
Google election ads	Synthetic or digitally altered images, video or audio that inauthentically depict real or realistic-looking people or events	The "Altered or synthetic content" checkbox; Google generates the label for Feeds, Shorts on phones and in-stream formats; in other formats the advertiser adds a clear and conspicuous disclosure (Google policy)	Inconsequential edits such as resizing, cropping, color and brightness correction, red-eye removal and background edits that do not create realistic depictions of events. Google warns at least 7 days before suspending for a violation
YouTube uploads (all creators)	Realistic content that makes a real person seem to say or do something they did not, alters footage of a real event or place, shows a realistic scene that did not occur, or uses AI-generated music as the main focus of the video	Answer the "AI use" question in YouTube Studio; labels appear on the player or in the description, and YouTube may add labels itself, including from C2PA data (YouTube help)	Unrealistic or animated content, beauty filters, color fixes, captions, upscaling, AI help with scripts or thumbnails, and cloning your own voice for voiceovers
TikTok	Realistic synthetic or manipulated media must be clearly labeled, by sticker or caption	A label such as "synthetic," "fake," "not real" or "altered" (TikTok guidelines)	No paid political ads or fundraising by politicians or parties at all. Synthetic media of private figures is barred, and synthetic media of public figures is barred when used for political or commercial endorsements or when it breaks another rule
X	Media presented as authentic that is significantly and deceptively altered or fabricated, or real media shared with false context, when likely to cause widespread confusion on public issues, affect public safety or cause serious harm	Enforcement depends on severity: limiting a post's reach, requiring deletion, temporary feature limits, or suspension for severe cases. The page as of October 2026 describes no user disclosure label for AI media (X policy)	Sharing manipulated media in non-deceptive ways is allowed; parody, commentary and fan accounts must be labeled as such in the account name and bio. X may take no action when it cannot tell whether media is misleading

Two things stand out. First, the platforms reach still images (Meta and Google expressly), which Idaho's statute does not. Second, the platform rules turn on realism, not on whether AI was used. A hand-made Photoshop composite that looks real needs the same checkbox as an AI image; an AI illustration that looks like an illustration usually does not.

Content Credentials and provenance

The Coalition for Content Provenance and Authenticity (C2PA) publishes an open standard, called Content Credentials, that attaches a record of a file's origin and edits, like a nutrition label for a picture or video. Its steering committee includes Adobe, Amazon, the BBC, Google, Meta, Microsoft, OpenAI, Sony and TikTok, among others. Some cameras, phones and editing tools add credentials, and YouTube says it may label content automatically when C2PA data shows it was made or altered with AI, a label the creator cannot remove.

For a campaign, credentials help two ways. A campaign that shoots real footage can keep credentials on its files so it can show later that a clip is authentic. And a campaign targeted by a fake can check whether a suspicious file carries credentials. But the absence of a credential proves nothing: many tools never add one, screenshots and many uploads strip them, and a credential says where a file came from, not whether what it shows is true. The watermarks and provenance lesson explains how the standard works and its limits, and the judging a claim lesson walks through checking a viral clip.

Worked example

Using AI honestly: practical house rules

Law and platform policy set the floor. A campaign's reputation depends on what it does above it, and the rules below cost nothing to adopt.

• Never synthesize a real person's voice or face saying or doing something they did not, including your own candidate, without a label, and never an opponent's at all.
• A human checks every fact in AI-drafted text against a primary source before it goes out. AI tools invent statistics, quotes and statute numbers that look right.
• Label realistic AI images, even where no rule requires it. "Illustration made with AI" in a caption is enough.
• Keep the originals. Save raw files, prompts and the dates assets were made, so you can show what is real if your own footage is attacked as fake.
• Do not let AI write testimonials, letters to the editor or reviews in other people's names. It is deceptive, and it breaks platform rules on impersonation and fake engagement.
• Write down who may use which tool, and which accounts hold the logins, as part of the security plan in lesson 10. An AI tool connected to the campaign email account can read the donor list.