The Pitch
A protocol for verifying the actions of autonomous AI agents on-chain using zero-knowledge proofs to ensure they operate within predefined safety bounds.
Problem
Currently, when an AI agent executes a transaction on behalf of a user, there is no cryptographic proof that the agent followed its system prompt or safety constraints. This limits the trust we can place in autonomous agents managing funds.
Proposed Solution
By wrapping inference in a TEE (Trusted Execution Environment) and generating a zk-proof of the execution trace, we can verify that the agent's actions were derived from the approved model and prompt before the transaction is accepted by the smart contract.
Why Blockchain?
Blockchain provides the immutable ledger for verifying the proofs, and smart contracts act as the programmable gatekeepers that only release funds if the zk-proof is valid.