The Agentic Governability Index
22 products, no two alike.

Grades name who owns the enforcement point: Green = full controllability, enforcement lives in the organization’s own control fabric (gateway, IdP tenant, own cloud account, MDM-authored policy, network, SIEM). Yellow = limited controllability, the vendor’s plane configured by the org. Red = no controllability, vendor-controlled. ∅ = security gap, no control surface exists. Click any well for the grade, the reasoning, and the tiered vendor documentation that proves it (T1 = vendor docs, T2 = vendor blogs and announcements, T3 = independent corroboration). Click a column to sort. Violet * = requires high engineering effort; the panel carries the detail. Amber ✱ = activation state: the grade is a ceiling, and the asterisk marks potential energy (opt-in, tier-gated, deployment-gated, build-required, preview, or gateway-dependent; the panel names which). On the A2A column, n/a means not applicable, and that is not a gap: single-principal tools have no cross-boundary agent surface to govern. ∅ elsewhere means a genuine security gap: no control surface exists. Guardrail wells carry a class tag: C = content safety, A = action approval. Click a Containment well to see which of the six layers reach that product: gateway credential, human identity, agent/NHI identity, product admin, network, endpoint.

Logging columns are deliberately split: RT Logging grades mechanism and latency (Green = stream or push arriving at a destination the organization controls, shipped or preview), Log Quality grades the four essentials (prompt/response content, agent actions, cost attribution, identity, human and agent; Red = one or more missing by any documented means). Data Boundary grades whether the org can enforce retention control, training-use exclusion, and residency, with its own levers where possible. Full rubrics, the T1 to T4 evidence policy, and the agentgateway baseline are in the methodology.