Observed and attested evidence
Developing
Configuration evidence remains separate from interview attestation
No. Application presence, permission scope, license assignment, activity, network traffic, data movement, and prompt inspection are separate evidence layers that support different claims.
An AI application name is a starting point, not a conclusion.
ChatGPT installed does not prove active use. An enterprise application registration does not prove broad adoption. OAuth consent does not reveal prompt content. A Copilot license does not prove Copilot is being used. A visit to an AI site does not prove sensitive data left the organization.
Different evidence supports different claims
Application inventory can establish presence in a source. OAuth evidence can describe permissions, consent, publisher context, owners, and assignments. License records can establish entitlement or assignment. Available activity records may support a usage clue within a stated time window. Network telemetry may identify traffic and transfer patterns. Deeper inspection products may observe prompt or MCP activity under a very different technical and privacy boundary.
Flattening those layers into “AI risk found” makes the conclusion sound stronger than the evidence.
Start with the relationship
Ask what was observed, where, when, and through which permission. Identify who owns the application, which access exists, which people or objects are related, what business purpose is claimed, and what remains unknown. Then decide whether the appropriate response is to allow, review, govern, educate, monitor more deeply, replace, or block.
The responsible finding is not “AI exists.” It is a specific, source-labeled condition about an application relationship that the MSP and client can investigate together.