AI Disclosure for Deepfakes

Deepfakes — realistic AI-generated or manipulated depictions of real people — face the highest transparency bar of any AI media. The EU AI Act specifically requires clear disclosure that the content is artificially generated or manipulated. Several US states add their own rules, particularly for elections and non-consensual content. A prominent, persistent on-screen label plus a written notice is the safe standard.

Real-world examples that require disclosure

A face-swapped video putting a real person's likeness into footage they never appeared in, a synthetic video of a politician "saying" something for satire or commentary, a voice-cloned audio clip attributed to a real person, and AI-manipulated video used in a legal or journalistic context all require prominent disclosure. Clearly labelled parody using an obvious impersonator (not AI-generated likeness) is a different category and is not treated as a deepfake under these rules.

Common misconceptions

• "Satire is automatically exempt from deepfake disclosure" — satirical intent does not remove the need for a clear synthetic-media label if the depiction is realistic.
• "A disclosure in the caption alone is sufficient" — for deepfakes specifically, a persistent on-screen label is expected in addition to a written note, because captions can be missed or cropped out when reshared.
• "Only political deepfakes are regulated" — the EU AI Act's transparency duty covers deepfakes broadly, not just political content, while US state laws often focus more narrowly on elections and non-consensual content.
• "Deepfake laws only apply to the original poster" — some jurisdictions also place obligations on platforms and, in certain cases, on those who knowingly redistribute unlabelled deepfakes.

Practical guidance on where and how to disclose

Burn a persistent label into the video frame itself (not just the platform caption) so it survives downloads and reposts, and repeat the disclosure in the written description. For likeness of real, identifiable people, consider whether you also need consent, since disclosure addresses transparency but not necessarily rights of publicity or defamation concerns, which are governed separately.

Current rules for deepfake content

• Apply a prominent, persistent on-screen label identifying the content as AI-generated or manipulated.
• Add a written disclosure in the description or caption as well.
• The EU AI Act mandates clear disclosure of deepfake content.
• US state laws add rules for election-related and non-consensual synthetic media — check local requirements.

Example disclosures

Frequently asked questions