Each boundary establishes a different kind of authority

BeaconGuard separates context, proposal, authorization, execution, and evidence so permission in one layer cannot silently become authority in the next.

BeaconGuard data-flow diagram separating context, AI proposal, exact-action authorization, execution, governing policy authority, authorization evidence, and independent verification into distinct trust boundaries.
Each boundary establishes a different kind of authority. Identity, delegation, and invocation access do not silently become permission for the final consequential side effect.

Boundary contracts

BoundaryEstablishesDoes not establish
Identity / delegationWho or what is acting, for whom, and within what boundsFinal permission for every consequential action
InvocationWhich MCP tool, A2A path, API, or application capability may be reachedAuthority for every side effect that capability can produce
AI / agent proposalThe exact candidate result or action to be evaluatedExecution authority
BeaconGuard authorizationALLOW or DENY for the exact material action under governing policyBusiness-system execution
Executor / system of recordValidated side effect and resulting authoritative business statePolicy authority
Evidence / verificationIntegrity, historical context, and reconstruction materialRuntime permission to execute

Core data flow

  1. Validate required identity, delegation, tenant/environment, freshness, and replay context.
  2. Admit the required invocation path; the AI or agent produces a proposed consequential action.
  3. Canonicalize the exact material action and resolve the current governing policy authority.
  4. BeaconGuard returns ALLOW or DENY and records authorization evidence.
  5. ALLOW may issue one-time execution authority bound to the exact action; DENY issues none.
  6. The executor recomputes the action identity, consumes valid authority once, and writes resulting business state.

Residual responsibilities

Related technical references