The perimeter assumed a human at every decision.
Thirty years of enterprise security drew a boundary: trusted people inside, everyone else outside, and the software between them as a conduit. That model presumes everything inside is authorised, and that a human stands wherever the business actually does something. An agent dissolves both assumptions at once — it is an insider that never sleeps, exercising authority it was granted once and uses a million times. A wall says nothing about what the system’s own hands are doing.
The established answer is to bolt governance onto existing platforms — a policy layer here, an audit trail there. A record a system keeps about its own decision is a claim, not evidence. Answerability cannot be added afterwards, because it is not a component: it is what the system stores at the moment it acts, and what it can show someone who was not there.
MIMUSE was built the other way round — the records, the authority and the proof are the foundations, and the products are composed on top of them.