Regulation is taking shape
The law has started to deal with faces and voices made by AI.
Below is what is in force and what is still only a bill, kept apart on purpose, each with a source you can check. This is a summary of published statutes and government material, not legal advice.
US federal
- In force
TAKE IT DOWN Act
Enacted May 2025; platform duties from 19 May 2026Covered platforms must provide a removal request process for non-consensual intimate imagery and take the image, plus known identical copies, down within 48 hours of a valid request. AI-generated and AI-altered images are covered. The FTC has begun enforcement.
Source: FTC - Bill
NO FAKES Act
Advanced by the Senate Judiciary Committee on 22 June 2026; not lawA bill that would create a federal right in digital replicas of voice and likeness, with a notice-and-takedown process. If it passes, the current state-by-state patchwork consolidates.
Source: Congress.gov
US state
- In force
ELVIS Act (Tennessee)
In force since 1 July 2024The first US state law to protect voice as a property right, aimed squarely at AI tools that clone a person's voice without consent. Signed 21 March 2024; 2024 Tenn. Pub. Acts ch. 588.
Source: Tennessee 주정부 - In force
AB 2602, AB 1836 (California)
AB 2602 from 1 January 2025; AB 1836 from 1 January 2026A contract that lets someone create a digital replica of a performer's voice or likeness must state specifically what will be made and obtain consent. Digital replicas of deceased personalities require the estate's permission.
Source: California Legislative Information
Korea
- In force
AI Framework Act (Korea)
In force since 22 January 2026Providers of generative AI must label output as AI-generated. Deepfake output requires a label a viewer can actually see. Fines are held back during an initial grace period.
Source: 국가법령정보센터 - In force
Unfair Competition Prevention Act (Korea)
In force since 8 June 2022Using a well-known person's name, likeness or voice for your own business, against fair commercial practice and to their economic detriment, is an act of unfair competition. This is the practical basis for commercial likeness and voice claims in Korea.
Source: 국가법령정보센터 - In force
Sexual Violence Punishment Act art. 14-2 (Korea)
Amendment in force since 16 October 2024Editing or synthesising a person's face or body into sexual material and distributing it is a crime; a 2024 amendment extended punishment to knowingly possessing, buying, storing or viewing such material.
Source: 국가법령정보센터
It is not only the law
The generation companies restructured first.
Below is what the largest generation services did in the past year with nobody requiring it. Once they signed with rights holders they had to account not for what they made, but for where it went and what was permitted.
- 3 September 2026
Suno
The AI music service capped downloads. On paid plans you can generate as much as your credits allow, but exporting is limited to 20 or 60 tracks a month with extra downloads sold separately. Its terms allow commercial use only of output taken through the official download path.
Generating and exporting became different products. Making is open; taking it out to sell is a permission and a record.
Source: Suno - November 2025
Suno, Warner Music Group
The lawsuit ended in a settlement and a licence agreement. The download-cap framework was made public in the same announcement.
The cap was not about server cost. A company that signs with rights holders has to be able to account for where its output went.
Source: Music Business Worldwide - 29 October 2025
Udio, Universal Music Group
The copyright suit ended in a settlement with recorded-music and publishing licences. The new service, due in 2026, is announced as trained only on authorised and licensed music.
A generation service now runs inside a licence. What it may put out is set by an agreement rather than by what the model can do. The settlement mechanics have not been disclosed.
Source: UMG, Udio - Since 2025
ElevenLabs
It runs a voice marketplace that pays the owners of the voices. By its own account creators have earned more than $22M. Famous and historical voices can be listed only by the person or their estate, with term, territory and channel set by contract.
A registry for voice, in everything but name: only the consenting are listed, the allowed scope is written into a contract, and use is priced.
Source: ElevenLabs
These companies acted in their own interest under the pressure of lawsuits and deals. None of it is an endorsement, and no law requires us. What the four have in common is a direction: making is open, while using someone's identity and taking the result out to sell become a permission and a record.
What this adds up to
Regulation moves first toward punishing and removing what has already been made: takedown duties, labelling duties, penalties. But removal alone never produces a permitted use.
For an AI service to use someone's face or voice, it needs somewhere to ask before it generates, and a record proving what was permitted. Today that address barely exists.
That is the position Rightis takes. A rights holder registers a face, voice, image or persona and sets the allowed scope; an AI service checks before it generates; permitted use becomes a contract and a log.
This page summarises published statutes and government material and is not legal advice. Consult a lawyer on any specific matter. Current as of 10 September 2026.