DoNotTrain
Enterprise infrastructure for creative-IP rights: register a work's cryptographic fingerprint, declare it off-limits to AI training, and hold court-admissible, independently verifiable proof of prior notice, without the asset ever leaving your control.
Overview
The question is now board-level: as models train on everything within reach, organisations and creators have no defensible, machine-readable way to assert that their content is not training data, and no proof they ever said so. DoNotTrain closes that gap with cryptographic opt-out records anchored to an immutable public ledger.
A work is fingerprinted locally, a SHA-256 and a perceptual hash, and registered in a single transaction; the asset itself never leaves the owner's environment. The result is a timestamped, tamper-evident record that any third party, counsel, auditor, counterparty, can verify independently, with no trust placed in us or any single vendor. Every registration generates a portable, receipt-grade evidence bundle.
Integrity is built in. A perceptual-similarity gate pre-empts ownership disputes by blocking registration of near-duplicate works, while a content-authenticity layer reads C2PA Content Credentials alongside an AI-generation classifier to keep synthetic media out of the human-creator registry.
Capabilities
- Cryptographic, court-admissible proof of opt-out and prior notice
- Immutable, independently verifiable record, no trusted intermediary
- Privacy-preserving by design: assets never leave the owner's environment
- C2PA content-authenticity + AI-generation gating for registry integrity
- Perceptual-similarity ownership protection against duplicates and squatting
- Portable, receipt-grade evidence bundle for every registration
Next project