You can spot it by what happens next. A Copilot agent goes
live and reads files that only a small circle was ever
supposed to see — but that, technically, far more people
have had access to for years. An agent acts “on behalf of
the organization,” but nobody decided up front who's
accountable when it gets something wrong — the agent
identity was never settled. Sensitivity labels, which mark a
document as public, internal, or confidential, have often
been in the tenant for years, but aren't enforced or are
applied incorrectly, so Copilot can't rely on them. A
misconfigured agent reaches half the organization within
hours before anyone notices — its blast radius, the
worst-case reach, was never limited. And teams across the
business build their own agents and hook them up to external
systems through MCP servers — connections based on the Model
Context Protocol, which let an agent reach tools and data
outside Microsoft 365 — with no central oversight.
Then there's a question that has nothing to do with your own
tenant: the EU Data Boundary, Microsoft's commitment to
process EU customers' data inside the EU, determines where
Microsoft may process your data in the first place. It
doesn't replace your own governance for access inside the
tenant, though. One covers where the data lives; the other
covers who gets to see it.