SELECTED WORK / 02 — TRUSTED AI
Agent Proof
As intelligent systems take action, make authority, activity, and responsibility easier to understand.
The question
When an AI system acts on someone’s behalf, a useful result is only part of the story. People also need to understand who authorized the action, what boundaries applied, and what happened along the way.
Agent Proof asks how those questions can become easier to answer in the workflows where AI is used.
What we are developing
Our work focuses on making AI permissions and activity more accountable. It explores the connection between authorized intent, action, and evidence people can review.
- Clarify authority. Establish what a system is permitted to do and where further approval is needed.
- Make activity reviewable. Preserve useful evidence of the actions taken and the permissions involved.
- Keep responsibility visible. Make clear who can review, intervene, and decide what happens next.
An operations team wants an AI assistant to prepare a purchase request. The assistant may gather information and draft the request, while a person retains authority to approve spending.
The important questions are concrete: was the action within scope, was approval required, and can the team review the sequence afterward? This example illustrates the accountability questions guiding Agent Proof.
Who it is for
Organizations exploring AI workflows, teams responsible for deployment and oversight, and builders who need to make authority understandable to the people relying on their systems.
Start with one workflow.
Tell us what you want an AI system to do, who it acts for, and where human approval matters.