Before an AI agent acts: a shared contract for intent, provenance and evidence
I maintain PIC Standard (Provenance & Intent Contracts), an Apache-2.0 protocol for making AI agent actions checkable before they happen. The idea When an agent wants to do something with consequences (send a payment, export personal data, delete resources), PIC asks it to first propose that action in a structured form: its declared intent, the impact class of the action, where the inputs came…
The PIC Standard (Provenance & Intent Contracts) is an Apache-2.0 protocol that ensures AI agent actions are checkable before they take place. When an agent wishes to perform an action with consequences, such as sending a payment or deleting resources, PIC requires the agent to propose the action in a structured format. This proposal includes the agent's declared intent, the impact class of the action, the provenance of the inputs, the claims the agent relies on, and the evidence supporting those claims.
A verifier then checks the proposal against the contract, which can result in allow, block, or require-more-evidence decisions. Guard integrations enforce the decision before tool execution, while applications using the HTTP bridge must enforce the returned decision. The process runs locally, with trust roots under the operator's control, and cryptography is not the focus but rather a supporting element.
PIC aims to provide a shared protocol for agents built in various languages and frameworks, allowing for unified safety checks and agreement on action contracts. The project is open to contributions, with first-time contributors encouraged to focus on documentation, while more experienced developers can work on TypeScript evidence mode, Go or Rust verifiers, or security reviews.
The project is currently at version 0.9.0, with the next milestone being a cross-implementation release.
Written by urgent.news from Dev.to's reporting — not their text. Machine-written — may contain errors; check the original before relying on it.
