# Observation over self-report

AgentShield AI Defense measures how AI systems read the web by recording their requests, not by asking models what they know.

A model describing its own knowledge is the system under test giving evidence about itself. This site treats such statements as claims and relies instead on observed requests and on coined strings published at recorded instants.

Every page here is served identically to every client. Nothing is cloaked and no content differs by user agent.

---

Reality marker: `asd-masanupo-naziduzi-b888a0`
Published: 2026-07-25T10:35:25.584Z
Source: https://agentshieldaidefense.com/probe/data.md

This marker is a coined string that exists nowhere else. Its later appearance in
a language model's output is observed evidence that this page was ingested.