How zero-knowledge proofs let an AI agent transact on behalf of a human — proving it has the right to act, without revealing who is behind it.
A zero-knowledge proof lets one party — the prover — convince another party — the verifier — that a statement is true, without disclosing anything beyond the truth of the statement itself.
Proving you have the right key to a locked room by opening the door — without ever showing the key.
Counterparties learn exactly what the proof attests to, and nothing else. The principal — the human on whose behalf the agent acts — is never identified by name, account, or history.
Fully homomorphic encryption lets an agent compute over sealed data without exposing the underlying inputs.
| Dimension | API Key | OAuth 2.0 | ZK Agent Identity |
|---|---|---|---|
| Identity model | Shared secret. Anyone with the key is the agent. | Delegated token bound to a user account. | Proof bound to a policy; the holder proves the right, not the identity. |
| Disclosure to counterparty | None — key is opaque, no claims attached. | High — user identity, scopes, and provider metadata are exposed. | Minimal — only the policy-scoped claims this transaction needs. |
| Revocation | Rotate the key globally — affects every caller. | Server-side token revocation list per provider. | Proof is self-contained and expires; revocation is a one-bit signal. |
| Audit posture | After-the-fact log scraping; no proof of intent. | Provider logs the who, what, and when — but the user is identifiable. | Cryptographic proof of authorization at the moment of action; principal remains pseudonymous. |
| Replay / theft risk | High — stolen key = full impersonation. | Medium — bound to client + redirect URI, but token theft works. | Low — proofs are single-use or tightly scoped; keys are never sent. |