Agent Trust: reference architecture for the governance of SI agents (AT-ARCH-001)

Organizations deploy autonomous SI agents faster than they can govern them: many cannot say how many agents operate in their infrastructure, who is responsible for each one, or what authority each one holds. Agent Trust is a neutral governance layer, independent of any single vendor or runtime, that gives every agent a verifiable identity, an accountable owner, declared limits, enforced policy, a continuously evaluated trust profile and a verifiable record of what it did. This document discloses its reference architecture: the agent sources it governs, eight platform capabilities, five core engines (Enterprise Agent Repository, Enterprise Agent Knowledge Graph, Trust Engine, Policy Enforcement, Governance Intelligence), the ecosystem integration and data layers, six functional views including a human-approval flow for high-risk operations, and a five-phase roadmap from v0.1 to v1.0.