Celebrity AI Voice Cloning in 2026 — What's Legal, What's Banned, What Apps Actually Allow

Tennessee's ELVIS Act (HB 2091/SB 2096), signed into law March 2024 and effective July 2024, makes it illegal to use someone's voice without authorization in AI-generated content if used commercially. Tennessee is the first US state with explicit AI voice cloning legislation. California's AB 2602, signed September 2024, requires consent for digital replica use of performer voices in entertainment work. New York's S7137 (2024) covers similar ground.

The federal layer is meaningfully behind. The NO FAKES Act (Nurture Originals, Foster Art, and Keep Entertainment Safe Act) was reintroduced in the Senate in March 2025 with bipartisan support but had not passed as of May 2026. The FTC has used existing deceptive-practices authority to pursue several voice-cloning fraud cases since 2024, but the underlying legal authority is patchwork. The European Union's AI Act, fully effective August 2026, includes deepfake disclosure requirements that will affect voice cloning by extension.

ElevenLabs, the leading voice-cloning platform, requires explicit voice consent through their Professional Voice Cloning program for any cloned voice, and prohibits unauthorized celebrity voice cloning in their terms of service. They maintain a public denylist of voices that cannot be cloned even with apparent consent (presidential candidates, certain protected celebrities).

Most AI girlfriend apps — Candy AI, DreamGF, Kupid, Muah AI — offer voice modes but do not allow named-celebrity voice cloning. Their voice systems are built on ElevenLabs or similar third-party voice infrastructure that filters out celebrity voice requests. What they do allow: voice tone customization (warm, low, breathy, energetic), accent selection (American, British, Australian, regional variations), and pacing controls. This covers most user intent without crossing the named-celebrity line.

Google Trends US shows steadily-rising search volume for queries like 'AI Taylor Swift voice,' 'AI Drake voice generator,' and 'celebrity voice AI app' across 2024-2026. The peak was the spring 2024 'fake Drake' AI track moment; the curve has normalized to a higher baseline since. What's measurable in the search data is that the bulk of the demand is for entertainment/parody/meme creation rather than for any darker use case — but the legal exposure for both is similar under the new state laws.

For users specifically interested in the AI voice experience without the legal mess, the AI companion apps' voice modes are the cleanest path. You don't get a named celebrity; you get a voice profile you can tune to feel like the archetype you wanted. A 'warm, low, Southern-accented female voice in her thirties' captures most of what someone searching 'AI Reba McEntire voice' actually wanted, with zero state-law exposure.

Taylor Swift's January 2024 deepfake images on X drove the platform to ban searches for her name for several days — covered in our [deepfake retrospective](/trending/taylor-swift-deepfake-x-ban-january-2024-retrospective). The fake-Drake AI track 'Heart on My Sleeve,' released April 2023 and pulled across streaming services within 48 hours, was the moment the music industry recognized the threat to artist economic rights. Scarlett Johansson's May 2024 dispute with OpenAI over the GPT-4o 'Sky' voice (which Johansson said sounded 'eerily similar' to her own voice) crystallized the consent question for major celebrities.

In 2025-2026 the cycle has continued: Adele's voice in unauthorized AI generations during her Vegas residency; Eminem's June 2025 legal action against an AI voice marketplace; Beyoncé's team's reported audio fingerprinting initiative to detect unauthorized clones. The pattern is consistent — celebrity discovers unauthorized clone, legal team responds, platform pulls content, the underlying capability remains widely available on open-source tools.

The legal pressure cycle does not affect the open-source layer of voice cloning meaningfully. Tools like XTTS, RVC (Retrieval-based Voice Conversion), and Coqui's open variants are freely available, run on consumer hardware, and produce results that approach commercial-platform quality. A user with a clean 60-second voice sample and a free weekend can produce a working voice clone of essentially any public figure.

This is the reason platform-level enforcement (ElevenLabs banning celebrity voices, Spotify pulling AI tracks) functionally raises the friction for commercial misuse but doesn't eliminate it. The honest read on 2026's voice-cloning landscape is: legal lines are now real for commercial use, but the underlying technical capability is widely distributed and effectively impossible to put back in the bottle. State laws are racing to address economic harm, not to ban the underlying capability.

Our prediction for the rest of 2026 and 2027: the NO FAKES Act passes (high probability given bipartisan industry support); the EU AI Act compliance enforcement kicks in (certainty given the August 2026 effective date); platform-level enforcement gets stricter (Meta, TikTok, X all rolled out tighter AI content disclosure requirements through 2025-2026); and the open-source layer keeps running unchanged in the background.

For users who came here looking for the AI voice experience without the legal mess, the recommendation stack is unchanged: use the major AI companion apps' built-in voice modes, customize tone/accent/pacing to capture the archetype you wanted, and don't ask the platforms to clone named real people. The substitution works cleanly. The archetype gets you 80% of what you wanted; the legal exposure goes to zero.