Authorize consequential AI decisions before execution.
TracePoint evaluates whether a proposed action is inside delegated enterprise authority. It applies policy, returns ALLOW, DENY, or ESCALATE, and binds permitted execution to the decision that authorized it.
An agent can have access without authority.
A tool permission can prove that an agent may reach a system. It cannot prove that the agent is authorized to make a specific commercial, financial, operational, or customer decision. TracePoint evaluates that business boundary before the target system changes state.
From proposed action to accountable outcome.
Connect the AI actor to an accountable enterprise principal.
Represent the proposed action, target, purpose, operation, and context.
Test authority, limits, policy, validity, and review requirements.
Return ALLOW, DENY, or ESCALATE with machine readable reasons.
Issue and verify a single use authorization bound to the exact action.
Connect execution, receipt, exceptions, outcome, and recovery.
A reusable language for delegated machine authority.
These objects keep identity, institutional accountability, action semantics, authorization, execution, and recovery connected without making TracePoint the identity provider or system of record.
AI actor
The externally identified agent, workflow, or service proposing the action.
Accountable principal
The enterprise role or person under whose authority the actor operates.
Authority envelope
The actions, resources, purposes, limits, validity, and review rules delegated to the actor.
Action descriptor
A system independent description of the exact action, target, purpose, operation, and context.
Decision record
The evaluated inputs, pinned policy and authority versions, outcome, reasons, and state changes.
Decision authorization
A server controlled, expiring, single use authorization bound to one approved action.
Action receipt
The external execution result connected to the originating decision and authorization.
Recovery context
The reversal, compensation, reconciliation, and exception path connected to the outcome.
Allow, deny, or escalate with explicit reasons.
ALLOW permits a bounded action. DENY stops an unauthorized action. ESCALATE routes a defined exception to an accountable reviewer. An approval queue is one escalation mechanism, not the product category.
Bind the proposed consequential action to its target resource, declared purpose, requested outcome, typed context, and validity window.
From accessible systems to authorized decisions.
See the authority gap. Then close it.
Actor and accountable principal are resolved
Authority Envelope bounds purpose, scope, and limits
Policy returns allow, deny, or escalate before execution
Connector enforces authorization and returns an Action Receipt
The decision travels with the action.
TracePoint currently creates a stable server controlled authorization fingerprint that binds the approved decision to its execution context. The authorization identifies its issuer and audience, targets one resource and operation, expires, and is consumed after use.
Execution produces an Action Receipt. Recovery context remains linked to the original Decision Record so reversal, compensation, or reconciliation does not become a disconnected process.
Operational change first. Decision authorization across the enterprise next.
Operational Change Authorization is the first working Control Pack. It verifies an identified service operations agent, delegated scope, destination, purpose, playbook, change limit, and rollback requirement before a local connector may act. Additional workflows require their own authority model, Control Pack, connector, and deployment validation.