Know whether an AI learned your work. And from which copy.
Every copy of a film or a piece of music receives its own invisible signature on delivery. If an artificial intelligence learns the work, it learns that signature too, and we know how to read it.
One film · six copies · six signaturesSignature 04 read in an AI model
The problem
AI models are learning from your work. Nobody can prove it.
An opt-out nobody can check
Rights holders have opted out of AI training. Without a technical means, there is no way to show that the opt-out was ignored.
Licences with no oversight
When a catalogue is licensed to an AI company, nothing lets you check how it is really used, whether the scope is respected, or whether the works are removed when the deal ends.
Leaks with no source
A film goes out in dozens of copies. When one of them resurfaces, online or inside an AI, nobody knows which channel it came from.
How it works
Three stages, from delivery to proof.
1 · On delivery
Sign every copy
An imperceptible signature, unique to the recipient, is embedded in the master. Picture and sound stay untouched.
2 · In the register
Record who received what
Each copy is logged in a dated, sealed register, along with the author's choice regarding AI.
We analyse the suspect model and hand over a dated report: did the AI learn the work, and from which copy?
What we deliver
Tools to prove, monitor and license.
ProofThe training report
A dated, reproducible report, checked by a person, stating whether an AI learned a signed work and through which copy. Built to support legal action.
LicensingThe licence audit
For collecting societies: how licensed works are really used, whether the agreed scope is respected, whether works are removed when the deal ends, and a basis for sharing revenue between authors.
CatalogueThe register and monitoring
All your signed copies and their recipients, plus ongoing monitoring of published AI models.
AI developersConsent before training
Check, before training a model, whether a piece of content is signed and what its author has chosen.
BroadcastLive signing
For broadcasters: a signature designed to follow a real-time stream, channel by channel.
MusicSound too
The same approach for music: an inaudible signature, unique to each copy, that we can read inside a music AI.
Who it's for
For those who create, own or distribute creative works.
Collecting societies
Producers and distributors
Festivals
Post-production labs and delivery providers
Broadcasters and platforms
Record labels and music publishers
AI developers who take copyright seriously
Our commitments
Discreet technology, built for rights holders.
ImperceptiblePicture and sound quality preserved, measured against industry standards.
Built for the lawCriteria set before each analysis, reproducible measurement, human review.
ConfidentialAI analysis stays on our servers. Your masters never leave your hands.
FrenchDeveloped and hosted in France.
ProtectedA patent application has been filed in France.
AssessedAn independent assessment is being prepared.
Contact
Let's talk about your work.
A demo, a trial on a few copies, a legal or technical question: drop us a line.
