How Sourcemark works
Sourcemark connects creator declarations with dataset checks, so AI teams can see what is restricted, what is available by licence and what needs review before training.
Declare, check and act before training.
Sourcemark connects creator and catalogue declarations with AI dataset checks. Creators and catalogues put an AI training choice and licence route on record. AI teams can then check candidate data before training or fine-tuning and see the declared status, licence route and next action.
Register
Creators or catalogues add the work they want to register.
Fingerprint
Sourcemark creates matching signals so the work can be recognised later.
Check
AI teams check candidate data against Sourcemark before training or fine-tuning.
Act
Results help route the next action: request a licence, exclude, hold for review or proceed under internal policy.
For creators and catalogues
A simple step before or after you publish.
Select the work, set your AI training choice and add the right licence route. Sourcemark handles the fingerprints, records and checkable signals behind the scenes.
Select the work
Choose the file, post, document, video or catalogue item you want to register.
Set your AI training choice
Mark it as available by licence or unavailable for AI training.
Add the licence route
Choose where requests should go: you, your representative, stock library, CMO, agency or publisher.
Create a checkable signal
Sourcemark creates matching signals for your work, including fingerprints and similarity-based signals where supported. These help AI teams check candidate training sets against Sourcemark and route the next action.
Every registration creates a timestamped record of what was declared and when. If your position changes, Sourcemark keeps a timeline of declarations rather than overwriting what came before.
For AI teams
Check candidate data before training.
AI teams can check candidate data against Sourcemark before training or fine-tuning. Matched works can return declared status, licence route, match type, declaration timestamp and a check record.
For unmatched files, Sourcemark returns no registered signal. A no registered signal result does not mean permission exists.
Prepare for checking
Register interest in Sourcemark checks for batch audits, workflow integration and dataset review.
Run checks across candidate data
Generate or submit supported fingerprints or matching signals and query Sourcemark before training or fine-tuning.
Interpret the result
Matched works can return a declared status, licence route, declaration timestamp and record details. Unmatched works return no registered signal.
Act before training
Route licence requests, exclude unavailable works, hold uncertain matches for review and document the checks carried out.
What a check can return
A Sourcemark check should not just return a match. It should return a result your team can act on.
Match result
Whether a registered work was found, including exact or similarity-based match signals where supported.
Declared status
Whether the work is available by licence, unavailable for AI training, not yet declared or needs review, with the timestamp and version of the declaration where available.
Licence route
Where to request permission when work is available by licence: creator, representative, agency, stock library, CMO or publisher.
Next action
Request licence, exclude, hold for review or proceed under internal policy.
Audit and governance record
A record of what was checked, what matched, what declaration was returned and when the check was made.
No registered signal
For unmatched files, Sourcemark returns no registered signal. Absence of a match does not mean permission exists.
What gets recorded
A Sourcemark record can include the information needed to make a creator's AI training choice checkable.
- Fingerprints and matching signals
- Declared AI training choice
- Timestamp and version history
- Licence route
- Declarer details, where public
- Verification page, where enabled
- Record of changes over time
- Cryptographic anchor for the declaration
Each Sourcemark is timestamped, versioned and cryptographically anchored, creating a tamper-evident record that can be checked later.
Sourcemark records declarations about files. It does not verify ownership, rights or authority.
Works alongside Content Credentials
Where Content Credentials are present, Sourcemark can read and preserve relevant provenance data. Content Credentials help show how a file was created or edited. Sourcemark records the AI training choice and licence route so they can be checked later.
C2PA is optional and is not required to create a Sourcemark record.
What Sourcemark does not do
Sourcemark creates a clearer record and checking layer before training. It does not stop scraping, remove content from the internet, verify legal ownership, clear rights, set prices or guarantee compliance.
Today, Sourcemark does not negotiate licences on your behalf. Future tools may support additional licensing workflows, including ways to present works that are available by licence and help route enquiries more directly.
Creators, catalogues, AI teams, legal teams and licensing teams stay responsible for final decisions.
Make AI training choices checkable.
Creators and catalogues can put choices on record. AI teams can check before training and route the next action.