Machine execution with a verifiable receipt.
An AI agent runs inside a trusted execution environment. The provenance layer binds the approved model identity, policy version, input commitment, tool permissions and output commitment into an attestation. A zero-knowledge proof can then verify selected claims without exposing proprietary weights or private prompts.
Model + agent binding
Register model hashes, runtime identity, policy scope and authorised machine wallet.
Permissioned runtime
Record tool calls, state transitions and human approvals inside a controlled enclave.
Selective disclosure
Publish proof and commitments to PoP L1 while retaining sensitive inputs and weights.