Is voice cloning legal? Yes, with consent.

Federal: the NO FAKES Act isn't law

The NO FAKES Act would create a federal right in your voice and likeness, with a notice-and-removal system for unauthorized digital replicas. The 2026 version (S. 4591, with H.R. 8915 in the House) passed the Senate Judiciary Committee unanimously on 18 June 2026, but neither chamber has passed it as of October 2026.

The FCC: AI voices in robocalls need consent

In February 2024, after a cloned Biden voice was used in New Hampshire robocalls, the FCC ruled that AI-generated voices are "artificial" voices under the Telephone Consumer Protection Act. Calls that use them need the called person's prior consent, like any prerecorded call.

Tennessee: the ELVIS Act

Since 1 July 2024, Tennessee protects a person's voice, real or simulated, against use without consent. It also reaches anyone who knowingly makes available a tool whose primary purpose is producing a particular person's voice without authorization. Violations can be civil and criminal.

California and the rest

California's right of publicity has long covered voice (Bette Midler won against Ford in 1988 over a sound-alike singer in an ad). Since 2025, AB 2602 voids contract terms that allow a performer's digital replica without a reasonably specific description of the uses, and AB 1836 requires an estate's consent for digital replicas of dead performers. Many other states protect voice through right-of-publicity laws, and more pass AI-specific rules each year, especially for election ads.

The AI Act: marking and disclosure

Article 50 of the EU AI Act has applied since 2 August 2026. Providers of AI systems that generate audio must mark it in a machine-readable way, so it can be detected as AI-made (systems already on the market before then had until 2 December 2026). People who publish a deepfake, including a voice that resembles a real person, must disclose that it's AI-generated, with a lighter duty for evidently artistic, satirical or fictional work. The Digital Omnibus that delayed other parts of the Act left these rules alone.

GDPR: a voice is personal data

A recording of someone's voice is personal data, and a voiceprint used to identify them counts as biometric data. Cloning another person's voice from their recordings means processing their data, so you need a legal basis, which in practice means their consent.

Denmark: copyright in your own voice

Denmark's bill L 49, introduced on 8 October 2026, would ban making realistic digital imitations of a person's voice, face or movements public without their consent, with exceptions for parody and satire. It's meant to take effect on 1 January 2027 if parliament passes it.

By ear, often you can't. Modern clones keep breathing, pauses and accent. Three things help more:

Check the marks. Under the EU rules, AI tools now watermark their audio, and several offer free checkers for their own marks: VoicedByMe's Check a clip, ElevenLabs' AI Speech Classifier, and Resemble's detector for Chatterbox. Each only recognises its own tool, and a determined person can strip a watermark, so a "not found" proves nothing.

Read the file's labels. Many tools write "AI-generated" into the file's metadata. Messaging apps usually strip it, so it helps with files, not voice notes.

For calls and voice notes, verify another way. If a voice asks for money or urgency, hang up and call back on a number you know, or agree a family code word in advance. That beats any detector.