Privacy model

Every stage of the pipeline is designed so that the party operating it learns as little as the stage permits. This page collects the whole picture; the per-product view for integrators is at Connect: privacy.

Knowledge by party

Provider

Learns that the user signed in, as in any session; not that a proof was generated at all.

Notary

Learns connection metadata and commitments; not session plaintext or disclosed values.

Keyring, during verification

Learns the disclosed fields, wallet, and policy; not session tokens or undisclosed fields.

Keyring, during issuance

Learns a blinded message and its validity proof; not the credential it is signing.

Chain observers

Learn address, policy, expiry, and a ciphertext; not identity, provider, or disclosed fields.

Regime authority

Learns the identity commitment behind a credential, on decryption; nothing beyond what the backdoor contains.

Three mechanisms carry the model

Redaction before disclosure. The TLSNotary presentation contains only the fields the proof plan names; the rest of the transcript exists solely as commitments. Nothing undisclosed can be extracted later, by anyone.

Blindness at issuance. The blind signature makes issuance and use unlinkable: even a complete record of Keyring’s issuance activity cannot be matched against on-chain credentials.

Escrowed identity. Anonymity is not absolute by design. The credential carries the holder’s identity commitment encrypted to regime authorities; decryption is possible for them alone and leaves everyone else’s view unchanged. The governance of that capability is a policy matter documented in the trust model.

The durable record

Only the chain persists publicly, and it holds no personal data. Everything more sensitive is either ephemeral (session data, disclosed fields during evaluation) or local to the user (trapdoor, blinding factors, raw session).