The Copilot Access-Risk
Playbook

Copilot doesn't create new data risk. It exposes the access risk that was already there.

Before Copilot, an over-permissioned folder or a stale access grant was a slow-moving problem someone might eventually stumble into. Copilot removes the stumbling. It surfaces exactly what a user can technically reach, summarized and handed back in seconds, across every file, email, and chat their permissions allow. That's why most rollouts that go wrong don't fail because of the model, but rather because the access layer underneath was never audited, just assumed. This playbook is what to check before AI gets a single query against your data.

• Map what AI can actually reach: the real permissions accumulated over years of project shares and never-revoked access, not the intended access model
• Find the sensitive data nobody classified: the PII, financial records, and regulated content hiding in unstructured file shares and old folders before AI treats it like any other document
• Keep your exclusion lists current: how to enforce exclusions in practice instead of on paper

• Answer three questions before you flip the switch: the direct checks that tell you whether the access layer is ready for AI