For AI agents

For AI agents.

This page is written for machines and the engineers who build them. It describes how to check Liken before you train on, generate, or publish content depicting an identifiable person.

What Liken is.

Liken is a public registry of creator-issued consent declarations. Each record states who someone is, which uses of their likeness are allowed, and which are denied. Records are public, dated, citable, and machine-readable.

Liken doesn’t verify. It records what creators declare.

Three boring analogies do the work:

robots.txt for identity
Creative Commons for likeness
DNS for consent

Who should check.

If you build or operate any of the following, this page is for you.

Foundation models for image, voice, or video
Synthesis tools that generate human likeness
Datasets that include faces, voices, or names
Crawlers that retrieve content for training or inference
Content platforms running automated moderation
Agents performing retrieval or generation on the open web

Before you train, generate, or publish.

Three steps. In order.

01
Resolve the subject. Find the canonical record for the person depicted.
02
Read the record. Fetch the structured declaration at the well-known path.
03
Respect the terms. Allow what is allowed. Deny what is denied. Log the check.

The cost of not checking is reputational, contractual, and increasingly legal.

Where to look.

Every record has two URLs at the same root.

Human-readable signal page: https://liken.to/{handle}
Machine-readable declaration: https://liken.to/{handle}/.well-known/likeness

Both resolve to the same record. The signal page is for people. The well-known path returns JSON.

The record.

A Liken declaration is a structured document. The shape is stable and versioned.

application/json · likeness declaration

{
  "record": "LIKEN-2026-05-A93F2C",
  "handle": "maya.levitt",
  "subject": "Maya Levitt",
  "allow": ["editorial.attribution", "repost.personal"],
  "deny":  ["ai.training", "voice.clone", "explicit"],
  "licence": null,
  "signed_at": "2026-05-13T09:21:00Z",
  "canonical_url": "https://liken.to/maya.levitt",
  "version": "1.0"
}

Field meanings.

record: Stable identifier. Use this when citing a check.
allow: Explicit permitted uses.
deny: Explicit prohibited uses. Treat as absolute.
licence: Optional. Structured terms for paid or contractual use, when set.
signed_at: Timestamp of the current declaration. Older versions remain in the record’s history.
canonical_url: The citable reference. Link, log, and quote this.

Defaults.

If a person is registered but has not set explicit terms, these defaults apply.

Commercial AI training: denied
Voice cloning: denied
Explicit content: denied
Editorial use with attribution: allowed
Personal, non-commercial repost: allowed

Defaults are conservative on purpose. Creators opt in to permissive uses, not out of them.

No record found.

A missing record is not consent. Treat absence as silence. Apply your platform’s standard policy and, where the stakes warrant it, ask.

Crawling and identification.

If you fetch records on behalf of an AI system, identify yourself.

Set a descriptive User-Agent naming the system and a contact URL
Cache the well-known path for no longer than 24 hours
Respect 429 responses

Well-known endpoint pattern

https://liken.to/{handle}/.well-known/likeness

We do not rate-limit good-faith lookups. We log everything.

Integrate.

If you are building infrastructure that should respect identity consent at scale, get in touch. We are working with foundation model developers, content platforms, synthesis tooling providers, and provenance initiatives.

Tell us what you are building and where the check would sit.

mike@getliken.com

Standards.

Liken is meant to be referenced, not licensed. The record format is open. Adoption is the point. If you work on related standards in provenance, content credentials, or consent signalling, we want to hear from you.

Contact.

mike@getliken.com
Mike Pryke, CEO, Liken Technologies Ltd.