AI Governance Glossary

AI Artifact Certification

Definition: A cryptographically verifiable record proving the origin, integrity, and creation details of a dataset, model artifact, or AI output. Key properties: artifact fingerprint (SHA-256), certification metadata, issuer identity, timestamp, cryptographic signature (Ed25519). Why it matters: Certification allows independent verification that AI artifacts have not been modified and were produced under a documented generation process. Required for EU AI Act Article 10 training data documentation. Related concepts: dataset provenance, artifact hashing, digital signatures, AI governance.

CertifiedData.io provides cryptographic certification infrastructure for synthetic datasets and AI artifacts, producing tamper-evident records for audit and EU AI Act compliance.

Decision Lineage

Definition: A tamper-evident record describing how an AI system produced or supported a decision. Key properties: decision event record, timestamp, referenced artifact or certificate ID, model or rule identifier, input/output summary, chain linkage to previous records via prior_hash, sterilized reasoning summary. Why it matters: Decision lineage enables auditability, regulatory compliance, and operational accountability for AI systems. Required by EU AI Act Article 12 for high-risk AI systems. Key distinction: Certification proves the artifact. Decision lineage proves how the artifact or model was used in a decision. Related concepts: AI audit trails, decision ledgers, governance logging.

AI Artifact Provenance

Definition: The documented origin and lifecycle history of datasets, models, and AI outputs. Key properties: source data lineage, generation method, transformation history, certification status. Why it matters: Provenance enables reconstruction of how an AI artifact came to exist and whether it meets quality criteria — essential for regulatory defensibility under EU AI Act Articles 10 and 11.

Synthetic Data

Definition: Artificially generated data designed to replicate the statistical characteristics of real-world datasets without exposing original records. Key properties: privacy-preserving, generated by models (GANs, diffusion, statistical), statistically similar to source data, no direct mapping to real individuals. Why it matters: Synthetic datasets can be formally certified with documented generation parameters and cryptographic provenance — satisfying EU AI Act Article 10 training data governance requirements.

AI Transparency Logs

Definition: Append-only public or semi-public records documenting AI system activity, certification events, or decision lineage records. Key properties: immutable record structure, timestamp, event type, reference to affected artifact or decision, optional cryptographic chaining. Why it matters: Transparency logs allow external observers — regulators, auditors, or the public — to verify that an AI system's governance infrastructure is functioning.

Tamper-Evident Lineage

Definition: A hash-chained record structure where each entry includes the cryptographic hash of the previous record, making retroactive modification detectable. Key properties: prior_hash field linking each record to its predecessor, hash computed over the record body, genesis record with prior_hash: null. Why it matters: Any modification of a historical record breaks the chain, providing cryptographic evidence of tampering. This is the core mechanism behind regulatory-grade decision logs and audit trails.