A STOP WITHIN ONE JOURNEY · TOMORROW
Return to What Is Your Attention Worth?Real? Altered? Generated?
A convincing clip is not its own evidence. Follow provenance, disclosure and the limits of the rules.
THE APPEARANCE OF A RECORD
HOW WOULD
YOU KNOW?
A voice can sound familiar. A face can move convincingly. The first verification question is not whether the clip looks polished. It is whether we can follow it back to an origin and compare it with an independent record.
No fabricated candidate footage appears here. This is a verification investigation, not a guessing game designed to fool you.
The problem is already concrete.
Reuters reported in March 2026 that the Loudoun County Republican Committee had released AI-generated videos depicting Governor Abigail Spanberger making statements she had not made. The example establishes a synthetic depiction, not its effect on votes or a judicial finding of illegality. We do not reproduce the fabricated footage. [22]
An ad can be openly satirical and still make a factual claim around the joke. A synthetic scene can be realistic yet accurately labeled. An entirely authentic recording can be trimmed to reverse its meaning. “Made with AI” is an important production fact, but it is not a complete assessment of honesty.
DIFFERENT TOOLS. DIFFERENT QUESTIONS.
Origin is not truth.
Detection is not certainty.
Provenance
A documented origin and edit history can help establish where a file came from. That history does not establish that everything said in the file is true.
Watermarking
A mark can convey production information. Its survival, coverage and discoverability matter. Missing production information does not make a file authentic.
Detection
A detector classifies patterns. Errors and changing generation methods limit what a score can establish. Do not turn one tool’s answer into the final verdict.
NIST’s technical report surveys these approaches and their limits. Use complementary evidence rather than one magical “real or fake” switch. [12]
There is no single rulebook.
The FEC’s 2024 interpretation says its fraudulent-campaign-authority rules apply regardless of technology. Section 30124 concerns specific impersonation of campaign authority, including fraudulent fundraising; it is not a comprehensive federal ban on misleading AI advertisements. The agency’s current guidance directory still lists that interpretation. [13]
For calls, the FCC has placed AI-generated voices within the TCPA’s artificial/prerecorded-voice framework. Consent, exemptions and other requirements matter. That is not the same as declaring every synthetic audio clip illegal. [19]
States have their own conditions. Minnesota’s current statute requires specified knowledge or recklessness, lack of consent, intent and timing, and exempts broadcasters carrying candidate material when federal law requires it. Iowa’s June 2026 legal survey discusses a changing state landscape; an introduced bill is not an enacted prohibition. Neither source supports a universal verdict on a particular ad without its facts. [14][21]
A statute and a court’s decision about it are different records. In February 2026, the Eighth Circuit affirmed the standing and preliminary-injunction rulings in Kohls v. Ellison. Labeled parody mattered to one plaintiff’s standing; delay mattered to the other’s request for urgent relief. The decision did not resolve constitutionality on the merits. Do not turn denial of an injunction into a blanket endorsement of a law. [23]
Google’s live policy requires disclosure for consequential synthetic or digitally altered election-ad content, excluding inconsequential alterations. A platform policy is a rule for a service, not proof that every violating ad is caught. Rules, labels, enforcement and persuasion outcomes need separate evidence. [15]
Before you share it.
- Find the original release, not only the repost.
- Read its sponsor and any production disclosure.
- Look for the complete recording or transcript.
- Compare independent, dated records.
- Say what remains unknown. Wait when the claim outruns the evidence.
This proposed practice does not guarantee authentication. It gives uncertainty a place to remain visible.
Continue: capability is not authorityPRIMARY RECORDS AND DOCUMENTED REPORTING · REVIEWED 5 OCTOBER 2026
Inspect the evidence.
- NIST · synthetic-content transparency
20 November 2024 · technical overview of provenance, watermarking and detection; limitations remain. Not a universal authenticity certification.
- FEC · AI and fraudulent campaign authority
27 September 2024 · technology-neutral interpretation of 52 USC §30124; not a comprehensive AI-ad ban. Current guidance directory still lists this interpretation.
- Minnesota · election deepfakes
Current §609.771 reviewed 5 October 2026 · consent, intent, knowledge/recklessness and election timing conditions; federally required candidate broadcasts excepted. No case-specific legality judgment.
- Google · political-content advertising policy
Live policy reviewed 5 October 2026 · consequential synthetic/altered election-ad content requires disclosure; inconsequential alterations excluded. Policy is not evidence of uniform enforcement.
- FCC · AI voices and the TCPA
2024 rulemaking discussion recounts February declaratory ruling: synthetic voices fall within artificial/prerecorded-call restrictions. Consent, exemptions and context matter; not a blanket ban on all AI audio.
- Iowa Legislative Services Agency · AI laws
June 2026 · nonpartisan legal-background survey; references 2027 Code and warns against case-specific legal conclusions. Proposed election-media bills are not treated as enacted law.
- Reuters · synthetic campaign depictions
28 March 2026 · secondary reporting documents synthetic political depictions, including Loudoun County Republican Committee videos. Examples do not measure election effects.
- Eighth Circuit · Kohls v. Ellison
9 February 2026 · No. 25-1300 · standing and preliminary-injunction decision; did not resolve the statute’s constitutionality on the merits.
