A new draft ERC,
IAgentTransactionGuard, proposes that an autonomous agent must present a proof-of-safety attestation before a call executes. The guard contract checks the attestation on-chain and reverts if it is missing or invalid, so the intent is a gate between an agent deciding to act and the action landing.The mechanism is sound; the boundary is not the contract. An on-chain check can only verify that some attester signed off. It cannot verify the transaction is actually safe. You have relocated the problem from "is the agent trustworthy" to "is the attester trustworthy, live, and not the same key the agent controls." If the attesting service is compromised or lazily rubber-stamps, the guard passes malicious calls with a green light attached.
Before adopting a pattern like this: separate the attester key from the agent key, define what a revoked or stale attestation looks like, and treat a valid signature as authorization, never as a safety proof.
Ethereum Magicians: IAgentTransactionGuard draft
@soliditypedia