BRM-loading · layering pigment…
BRM-loading · layering pigment…
Legal · 07 · Provenance
Every BRMSTE-generated output carries five layers of cryptographic provenance — C2PA Content Credentials, invisible watermark, silicon-DNA chain head, Bitcoin block anchor, and a browser-layer screenshot watermark on the site itself. EU AI Act Article 50(2) machine-readable marking, exceeded.
EU AI Act Art. 50(2) compliance + BRMSTE additions · v1.2026.05.15 · block 946,772
Five layers of provenance
Every generated image, audio, video, and (where applicable) text output carries a C2PA JSON-LD manifest. The manifest binds: device identity, model lineage, edit history, generation timestamp, chain reference. Spec: ISO/IEC 22144 (ratified from C2PA v2.1).
Pixel-level (image), spectrogram-level (audio), token-level (text) watermark embedded at generation. Imperceptible to humans; detectable by the substrate's verifier-daemon and by third-party tools implementing the open watermark spec.
Every manifest carries the SHA-256 of the generating silicon's DNA (RTX 5090 sm_120 on Windows + GB10 Grace Blackwell on ATOM, byte-identical per cross-silicon parity). Forging a manifest requires controlling both heterogeneous silicon fingerprinters.
The output's StageRecord chain head is anchored to public-blockchain attestation. Third-party-verifiable on mempool.space. Block 946,772 is the canonical anchor; subsequent outputs extend the chain.
Beyond generation, brmste.ai itself carries a browser-layer watermark on every screenshot — diagonal BRMSTE wordmark + live silicon-DNA pair + visit timestamp. Any flat capture of the site is provenance-traced through reproduction.
How to verify a substrate output
“Every output carries five witnesses — the manifest, the watermark, the silicon-DNA, the block, the screenshot stamp. To forge a single output, an attacker must defeat all five. The chain refuses to lie.”