Continuous Temporal Auditing of AI Training Data Disclosure: A Cryptographically Anchored Observational Framework at Ecosystem Scale

We present CROVIA, a continuously operating observational system for monitoring training data disclosure practices across the AI model ecosystem. Our system operates at ecosystem scale, monitoring 4,843 models from 1,152 organizations with automated daily observation cycles anchored to a cryptographic Merkle chain now at height 5,480. Analysis reveals that 57.6% exhibit CRITICAL disclosure gaps and the ecosystem-wide average compliance score is 26.8% across 20 disclosure elements mapped to 11 regulatory jurisdictions including the EU AI Act, GDPR, CCPA, and China's Generative AI Provisions. We introduce the Temporal Proof of Absence (TPA) — a cryptographically verifiable artifact asserting that specific disclosure elements were absent from a model's public documentation at a specific point in time, deployed across 200 models with full proof chains including Pedersen commitments on secp256k1, Ed25519 signatures, and temporal pressure scoring. Live registry: https://registry.croviatrust.com