RECEIPT VERIFICATION
VERIFY
WITHOUT
TRUSTING US.
Check AIR v1 and legacy receipts from confidential AI inference. This stateless surface validates receipt signatures and structure; API callers can add model, hash, freshness, platform, and attestation inputs for stricter verifier-backed reports.
Samples
Verifying...
Confidential AI verdict
Verification checks
Limitations
- Confirms the receipt is correctly signed and structurally valid.
- Can derive the receipt key from a supplied attestation document and check AIR attestation hash binding when the sidecar is supplied.
platform_attestedmeans the attestation is authentic and key-bound;tee_provenanceadditionally requires caller-supplied runtime PCR policy.- Does not prove data was deleted after processing.
- Does not constitute a compliance determination.
- Deployment-specific trust policy depends on expected measurements, model allowlist, and freshness inputs.
Privacy:
Uploaded receipts are processed in memory and are not persisted by the verifier application. Cloud Run request logs contain metadata such as path, status, trace ID, user agent, request size, and client IP; they do not contain request bodies and are retained for 30 days. This is not a proof of memory zeroization or deletion. No app analytics, tracking scripts, or third-party page assets are used. Source: github.com/cyntrisec/EphemeralML