AI voice disclosure
Direct answer
AI voice needs disclosure when it is realistic — especially cloned voices of real people or synthetic narration that could be mistaken for a real recording. Cloning an identifiable person’s voice also triggers consent and human review. YouTube’s altered-content disclosure explicitly covers synthetic voices; TikTok requires labeling realistic AI audio. Under the EU AI Act from 2 August 2026, deployers must disclose realistic manipulated audio resembling real persons or events.
Applies when
- Cloned voice of a real person.
- Realistic synthetic narration.
- Voice alteration making someone appear to say something they did not.
Doesn't apply when
- Obviously robotic or stylised voice where no one could assume a real recording.
Worked decision
Example
Generic synthetic voice-over on real footage
Disclosure recommended
The voice is synthetic; disclose it specifically to stay truthful and platform-safe.
Next steps
Required action
- Disclose synthetic voice in the platform flow and caption where sensible.
- Obtain written consent for cloned voices.
- Record voice model, scope and approver.
Publishing layer
Platform differences
Law, platform and client layers
EU AI Act Article 50(4) requires deployers to disclose realistic AI-generated or manipulated image, audio and video resembling real persons, objects, places or events. Use the platform’s native label and keep your own evidence record.
Evidence
Source
Freshness and accountability
Last verified
Applies from 2026-08-02 · Last reviewed 2026-09-10 · Current
Run this exact scenario
Relevant content, AI-use and platform inputs are preselected. Confirm the remaining facts before relying on the result.
Who is affected
- Podcast, social and video teams using synthetic narration.
- Agencies producing multilingual voice-overs.
- Legal teams handling voice-cloning consent.
More scenarios
| Scenario | Outcome | Why |
|---|---|---|
| Cloned voice of the CEO | Human review required | Identifiable person: consent, approval and wording need a human owner. |
| AI denoising of a recorded interview | No disclosure identified | Restoration does not change what was said; record it in the manifest. |
Practical marketing examples
- “Narration is AI-generated.”
- “Voice cloned with the speaker’s consent for accessibility.”
Related pages
Disclara separates binding law, platform policy, client policy, technical provenance and recommended wording. Information on this page is not legal advice. See editorial policy.