State of AI Identity Licensing 2026

A bounded primary-source review of the rules, proposals, contracts and technical standards shaping how a real person’s face or voice can be authorized, generated and verified in 2026.

Original research with a defined limit. This report codes 14 primary sources reviewed through September 5, 2026. It is not a global law census, legal advice or evidence that every source applies to every campaign.

14primary sources
5authority classes
8coded controls
0sources covering all controls

The finding

AI identity licensing is not becoming one rule. It is becoming an operating stack. Rights and contracts answer whether a person authorized a use. Advertising and synthetic-media rules answer what an audience must be told. Provenance standards answer what can be checked about a file and its history.

SocialGravity’s inference from this review: a buyer needs permission, disclosure and provenance as separate connected records. A synthetic label cannot supply consent. A licence cannot prove which output was generated. A valid provenance claim cannot decide whether the depicted person agreed.

What changed in version 1.1

This update adds the enacted TAKE IT DOWN Act as a narrowly scoped removal and remedy signal, adds SAG-AFTRA interactive-media guidance on written consent, intended use, compensation and suspension of new generation, and updates the technical standard to C2PA 2.4. The authority class and limits of each source remain explicit in the dataset.

What the sources address

These counts show topic coverage inside this dataset. They are not legal requirements, maturity ratings or performance scores.

Download the chart as SVG

Five forms of authority

5Enacted law or regulation
1Proposed legislation
4Government report or guidance
3Collective agreement
1Technical standard

Binding law, a proposed bill, collective bargaining, government guidance and a technical standard do not carry the same legal weight. The dataset keeps them separate so a reader can see whether a statement is a rule, proposal, contract pattern or implementation mechanism.

Five conclusions for buyers

1. Permission is becoming explicit and use-specific

9 of 14 sources address permission or a right to control use. 6 address contract form or specific scope. The pattern is strongest in the California contract rule, the proposed NO FAKES Act and SAG-AFTRA agreements, but the legal effect differs between them.

2. Asset access is its own risk

Only 2 sources in this bounded review directly address access, security, storage, transfer or destruction. That makes asset custody an easy operational gap even where the permission language is detailed.

3. Compensation and consumer disclosure answer different questions

4 sources address compensation or a paid relationship. 3 address audience disclosure. Paying the person does not make an endorsement truthful, and labeling an output does not establish that the person was paid or agreed.

4. Provenance is necessary but deliberately limited

3 sources address technical origin, marking or provenance. C2PA and NIST describe ways to bind, label or recover information about content. They do not convert a missing licence into permission.

5. Lifecycle control is moving into the contract and the credential

7 sources address duration, additional uses, termination, expiry, destruction or credential status. A buyer still needs an explicit rule for future generation and a separate rule for outputs already produced.

A working category vocabulary

These are SocialGravity’s operational definitions for this report. They do not replace a statute, collective agreement or signed contract.

TermWorking definition
Digital replicaA computer-generated representation readily identifiable as a real person’s voice or visual likeness.
AI identity licenceScoped permission for a named party to use defined identity elements for an agreed AI use, term and production route.
Consent of recordEvidence that the identified person agreed to participate and to the reviewed version of the proposed use.
Intended useThe buyer, campaign, script or claims, channels, territory, languages, formats and dates the permission covers.
Asset custodyThe controls governing who can access, copy, transfer, use or destroy captured face, voice and model assets.
Permitted generationA generation event allowed by the current licence, covered asset version and agreed production constraints.
Content provenanceVerifiable information about an asset’s origin and history. Provenance is not automatically proof of personal permission.
Output receiptA signed record connecting a released file to its generation event, licence references and content credential.
Lifecycle controlThe rules and technical actions for additional use, expiry, termination, revocation and existing outputs.

Use the underlying evidence

The operating implication

A defensible transaction needs more than one document or label. The practical chain is identity verification, reviewed consent, a signed licence, secured payment, terms-locked generation, a verifiable receipt and revocation control.

Inspect the evidence chainBring a licensing brief

This report is informational research. Applicability depends on the person, work, agreement, jurisdiction and use. Obtain qualified legal advice for a real transaction.

Common questions

What is the state of AI identity licensing in 2026?

The reviewed sources show separate movement around permission, use-specific contracts, compensation, disclosure, provenance and lifecycle controls. No source in this bounded dataset covers the complete operating chain.

Are synthetic-media labels proof of consent?

No. Disclosure tells an audience that content is synthetic or sponsored. Permission requires its own evidence from the person or authorized right holder.

Do Content Credentials prove an AI likeness is licensed?

No. Content Credentials can provide signed provenance assertions. They do not decide whether the depicted person granted the required rights.

Is the dataset a complete survey of global law?

No. It is a transparent, reproducible review of 14 selected primary sources through September 5, 2026. The methodology states what was included and excluded.