Choose the claim before you issue the proof
Certify synthetic data or notarize an existing AI artifact
Both workflows produce cryptographically verifiable records. Certification covers supported synthetic-data generation facts; notarization covers the integrity and existence of an artifact submitted from outside that generation workflow.
CERTIFICATION
Generate certified synthetic data
Use a supported CertifiedData generation workflow. The issued certificate can bind the generated dataset fingerprint, generation metadata, timestamp, schema version, and issuer signature.
- ✓Supported synthetic-dataset issuance workflow
- ✓SHA-256 artifact fingerprint
- ✓Ed25519-signed certification record
- ✓Machine-verifiable generation and issuance facts
NOTARIZATION / ATTESTATION
Protect an existing external artifact
Use the AI Artifact Notary for an existing model file, output, report, external dataset, or other artifact. The notary fingerprints the submitted bytes and issues a signed integrity/existence attestation.
- ✓Existing artifact submitted to the notary
- ✓SHA-256 artifact fingerprint
- ✓Ed25519-signed attestation record
- ✓No claim that CertifiedData created or validated the artifact's contents
Privacy parameters are not a privacy guarantee
A certificate can record generation metadata, including privacy-related parameters when present. An epsilon field, a DP-labelled mode, or a signed record does not by itself establish differential privacy. CertifiedData only makes a mathematical differential-privacy claim for a workflow that actually implements the required clipping, calibrated noise injection, and privacy accounting and binds the applicable mechanism into the evidence record.
Already have a proof?
Verification checks the signed payload, artifact fingerprint where applicable, proof-specific trust root, and current record status. It does not turn a valid signature into a broader legal, quality, privacy, or compliance conclusion.