V-TSAL: Verifiable Training Signal Attestation Layer — A Cryptographic Protocol for AI Training Data Provenance and EU AI Act Compliance

V-TSAL (Verifiable Training Signal Attestation Layer) is an open cryptographic protocol for AI training dataset attestation. It produces Training Signal Receipts (TSRs) — machine-verifiable records that prove an AI training dataset existed in a verified, forensically examined state at a specific point in time. The protocol uses BLAKE3 Merkle tree fingerprinting over dataset file structures, Ed25519 digital signatures issued from a Hardware Security Module, and a public transparency ledger. Any party including a regulator can independently verify a TSR using a single publicly available key without contacting the issuing authority. V-TSAL addresses the technical standard gap in EU AI Act Article 10 and Annex IV — which define what must be demonstrated about training data but specify no technical mechanism for producing machine-verifiable compliance evidence. This specification has been submitted to the EU AI Act high-risk classification consultation (May 2026).