EMILIA Protocol
An open protocol and gate that checks an AI agent's authority before it can complete a sensitive action, such as releasing a payment, and produces evidence that can be independently verified rather than trusting the agent.
EMILIA Protocol is an authority control plane for autonomous agent work. Its Gate component checks authority before a protected tool call is allowed to run, binding a human approval to the specific details of an action, such as a payment's amount, currency, vendor, and destination, so that changing any of those details after approval invalidates it. The resulting evidence is independently checkable against the verifier's own trusted keys and rules, rather than relying on trust in the agent itself.
It is designed to sit in front of a tool a team already uses, at the service that holds the real credential rather than only inside the agent's own process, so a bypass of the agent does not also bypass the control. Example integrations cover MCP and HTTP tools, Hugging Face's smolagents framework, and a GitHub merge check that can be made a required status check.
The project is explicit that it proves and prevents only on paths a deployment fully mediates, and that a valid approval does not itself establish that the underlying details, such as bank information, are legitimate. It is written in TypeScript under the Apache-2.0 license.