---
title: "The Agent on Trial | KYE Protocol™ blog"
description: "A successful AI-agent action is not a legitimate one. Put the action on the stand and ask the one charge a dashboard never answers — was it authorised? The verdict is Authority Finality™, dramatized in the Authority Court."
url: https://kyeprotocol.com/blog/the-agent-on-trial/
lang: en
source: "KYE Protocol"
---

> A successful AI-agent action is not a legitimate one. Put the action on the stand and ask the one charge a dashboard never answers — was it authorised? The verdict is Authority Finality™, dramatized in the Authority Court.

23 July 2026 · KYE Protocol™ · Doctrine · ― Is Not Authority

# The Agent on Trial.

Your agent took an action. It was fast, it was fluent, and it worked. None of that answers the only question a regulator will ask: was it allowed to?

[Enter the Authority Court](https://kyeprotocol.com/authority-court/) [See Authority Finality™](https://kyeprotocol.com/authority-finality/)

A successful action is not a legitimate one.

## What "successful" proves

Technical success answers "did the action run?" — and it answers it cleanly.

- The call completed without error and returned the expected result.
- The written policy compiled into code was satisfied.
- Access control let the caller reach the system it touched.

Each of those is real and worth having. But together they prove the action _could_ happen, never that anyone held the right to make it happen.

## Where it stops

Put the action in the dock and the defence runs out of exhibits fast.

"The policy owner approved it" — but owning the rule is not being the source of the mandate. "We passed our annual attestation" — but a point-in-time certificate is not a per-action approval. "Our access controls allowed it" — but reaching a system is not holding authority over it. "We have the full trace" — but a trace records what happened and stays silent on the grant behind it. Every argument answers a different question than the charge.

- Success is the floor, not the defence.
- A rule being satisfiable is not the authority to act on it.
- None of these exhibits contains the grant, or proves it was live when the action shipped.

## The charge

The court narrows to one question, asked of one action, at one moment.

Did this principal hold delegated authority for this purpose, within scope and limits, from a source empowered to grant it — and was that authority still live at t=0, the instant before the world changed? KYE Protocol™ answers exactly that at the action boundary and seals the basis into a replay-verifiable [Evidence Pack™](https://kyeprotocol.com/evidence-pack/) a regulator can check from public keys alone. That is the verdict, and it has a name: [Authority Finality™](https://kyeprotocol.com/authority-finality/). Read the full dramatization in the [Authority Court](https://kyeprotocol.com/authority-court/).

## Honest scope

KYE Protocol™ proves authority, not truth.

The trial does not decide whether the agent's answer was correct or its model well-calibrated — that is the work of evaluation, and it stays necessary. KYE Protocol™ proves exactly one thing, and proves it completely: whether the actor was authorised to take the action. The courtroom is a metaphor, not a legal proceeding — a way to picture the one boundary a dashboard never guards.

> A successful action is not a legitimate one.