Deepfake disclosure and likeness-rights clearance
ai-deepfake-disclosureDomain: ai-transparencyType: policyDescription
Deepfake regulation has the wrong-name problem. The statutes label themselves after the technique (deepfakes, synthetic media, AI-generated likenesses), but the operational obligation is almost entirely about the underlying right being violated. California AB 1836 is right-of-publicity legislation that happens to mention generative AI. Tennessee's ELVIS Act is the personality-rights statute extended to voice. EU AI Act Article 50 is a transparency requirement that triggers when output depicts a real person doing something they did not do. The work for an operator is therefore mostly the work of mapping each piece of synthetic output back to whichever underlying right (publicity, defamation, copyright, privacy, non-consensual intimate imagery) actually anchors the violation. A working disclosure and clearance flow has four pieces. Intake is the policy that decides which outputs require clearance before publication: any depiction of an identifiable real person, any voice clone, any deceased-person likeness used commercially, any sexually explicit synthetic content. Clearance is the rights workflow itself: consent capture from the living subject, estate or successor-in-interest authorization for deceased likenesses (AB 1836 expressly reaches estates for 70 years post-mortem, Tennessee's regime is similar), and a documented refusal path for content the platform will not run regardless of consent. Disclosure is the per-piece label on output the platform does publish, calibrated to the regulation: EU AI Act Article 50 requires clear and distinguishable disclosure for deepfakes depicting real persons; California's regimes require labeling on deceptive election communications and certain sexually explicit content. Takedown is the report-and-removal pipeline for content that escapes the clearance step, governed by the Take It Down Act's 48-hour window for non-consensual intimate imagery and by parallel state regimes for other categories. The deadlines and thresholds matter. AB 1836 makes commercial use of a deceased personality's AI-generated voice or likeness without estate consent civilly actionable, with damages set at the greater of $10,000 or the profits attributable to the use; it took effect 2025-01-01 and reaches conduct distributed in California. Tennessee's ELVIS Act has covered unauthorized voice-likeness use since 2024-07-01, with damages including treble for knowing violations. EU AI Act Article 50 transparency obligations on deepfakes apply from 2026-08-02. The Take It Down Act's 48-hour removal window starts when the platform receives a valid notice from the depicted person or an authorized representative. The labeling regimes and the removal regimes are on different clocks; the labeling clocks are calendar-anchored, and the removal clocks are notice-anchored. The genuinely awkward part is the cross-regime mismatch on what counts as consent. California AB 1836 wants written authorization from a specific class of estate-side rights-holders; Tennessee's ELVIS Act runs on the personality-rights frame; the EU AI Act runs on transparency rather than consent and tolerates disclosed deepfakes that would be actionable in the US right-of-publicity regimes. A clearance record that satisfies one regime may be insufficient for another, and the operational hedge most operators reach for is to require both the right-of-publicity authorization and the deepfake disclosure, treating each as a separate gating step rather than a single workflow.
Applicability
Applies when: features include synthetic-media.
Required by (4 regulations)
- California AB 1836
Prohibits the commercial production or distribution of a digital replica of a deceased personality's voice or likeness in expressive works without prior consent of the rights-holder; consent class includes the personality's estate or transferee for 70 years post-mortem.
California AB 1836 (2024), amending Cal. Civ. Code §3344.1; effective 2025-01-01
- EU AI Act
Article 50(4) requires deployers of an AI system that generates or manipulates image, audio, or video content constituting a deepfake to disclose that the content has been artificially generated or manipulated, with limited carve-outs for satire, parody, and law-enforcement use.
Regulation (EU) 2024/1689 (AI Act); deployer disclosure obligation under Article 50(4) applies from 2026-08-02
- Tennessee ELVIS Act
The Ensuring Likeness, Voice, and Image Security Act extends Tennessee's personality-rights regime to voice-likeness; unauthorized use of an individual's voice in a manner that would lead a reasonable person to believe the individual performed the work is civilly actionable, with treble damages for knowing violations.
Tennessee Ensuring Likeness, Voice, and Image Security Act, 2024 Tenn. Pub. Acts ch. 588; effective 2024-07-01
- TAKE IT DOWN
Covered platforms must establish a notice-and-removal process for non-consensual intimate imagery, including AI-generated depictions of identifiable individuals, with content removal within 48 hours of a valid report and reasonable efforts to remove identical copies.
Take It Down Act, Pub. L. 119-12 (2025)
Evidence formats
- Deepfake takedown SOP with 48-hour notice-to-removal target
- Deceased-likeness consent and estate-authorization policy
- Voice-likeness rights-clearance log with subject identifier + consent scope
- Per-piece deepfake disclosure copy / label specification
- Refusal-path policy for content the platform will not produce regardless of consent
- NCII reporting intake form and verification workflow