How we protect.

Secure Ingest

Your file is uploaded over TLS into an isolated, per-user storage vault. We never read its contents in plaintext on disk — only the byte stream needed to compute hashes.

ImplementationSHA-256 over the binary, perceptual hash (DCT-based pHash) for images, file metadata extracted server-side.

Fingerprint & Seal

The hashes are written to your provenance ledger alongside an irrevocable timestamp. This becomes your digital wax seal — proof you held the work first.

ImplementationRecords are owned by your account under RLS; only you (and a court order via subpoena to us) can read them.

Dataset Audit

We compare your fingerprints against sampled scraping corpora and a curated list of public datasets known to feed today's models.

ImplementationHeuristic v1 — exact-hash + perceptual-distance matching against LAION and Common Crawl samples. Not exhaustive; see Beta Promise.

Adversarial Defense

Future works can be released with an embedded poisoning layer. Imperceptible to humans, hostile to the gradient descent that trains generative models.

ImplementationGenesis Batch — request access from your vault. Effectiveness varies by model architecture and is documented per release.