Control Responsibility Matrix

This page helps reviewers understand which responsibilities BeaconGuard owns, which remain with the application and system of record, and which stay outside BeaconGuard’s control boundary. It is intended to reduce ambiguity during architecture, security, risk, and procurement review.

Responsibility model

BeaconGuard governs whether an AI interaction is allowed to proceed under policy. BeaconGuard does not replace transaction systems, systems of record, model providers, or downstream operational ownership.

Responsibility matrix

Responsibility BeaconGuard Application Model provider System of record Operations / reviewers
Authorization decision at the AI request boundary Owns policy evaluation and explicit allow/deny decisioning Submits normalized request context and enforces returned decision Does not own authorization policy Does not own AI request authorization Review boundary correctness and policy fit
Business transaction execution Does not execute business transactions Orchestrates allowed workflow steps Does not execute business transactions Remains authoritative for transaction/state recording Validate control separation and escalation paths
Model behavior and output quality Does not guarantee model quality or business correctness Determines how allowed outputs are used Owns model behavior and model-serving characteristics Does not determine model quality Evaluate suitability and downstream control expectations
Decision evidence and review context Emits structured decision evidence within its control boundary Correlates BeaconGuard decisions with workflow context as needed Does not provide BeaconGuard policy evidence May hold downstream business records, not BeaconGuard decision evidence by default Review evidence, trace decisions, and assess control operation
Fail-closed behavior when required conditions are not met Denies requests when policy or trust conditions are missing, malformed, or out of bounds Must respect and enforce deny outcomes Does not control BeaconGuard fail-closed semantics Not responsible for AI request gating Validate that failure handling is explicit and reviewable

What reviewers should verify

Next step

Reviewers can continue structured evaluation through the Reviewer Kit, Trust Center, and Contact path.