Execution Authorization Infrastructure

Authorize the consequence before it becomes real.

reht sits at the consequence boundary where business or agent intent can become real-world effect. It determines whether this exact proposed consequence is authorized here, now, under the current mandate, evidence and state.

Open reht Understand the category VALO Research
Canonical category

Consequence Governance governs the transition from intent to real-world effect.

Consequence Governance is the category. reht is the consequence-time authorization mechanism inside it. Governed Effect Path is the architecture. NO_DIRECT_EFFECT_PATH is the invariant that prevents any model, agent, tool, workflow or connector from bypassing that boundary.

Category

Consequence Governance

Governs whether a proposed consequential state transition may legitimately become real under current authority and conditions.

Boundary mechanism

Execution Authorization Infrastructure

The exact proposed consequence is cleared or refused using fresh authority, evidence and state.

Architecture + outcome

Governed Effect Path

Only valid authorization reaches effect. Trusted Execution is the evidenced downstream outcome, not the governance category.

Consequence boundary

Capability is not authority. Access is not authorization.

A model may be capable of an action and an identity may have access to a system without being authorized to perform this action now. Consequence-time authorization applies the current mandate and evidence to the exact proposed consequence.

Governance inputsPurpose, mandate, delegation, policy, limits and accountability.
EvaluationCurrent evidence, uncertainty, risk, context and state are assembled.
rehtThe exact proposed consequence is cleared or refused at the consequence boundary.
RACSThe deterministic authorization outcome is expressed for enforcement.
PEP / GatewayThe authorization is enforced against the action that is actually executed.
VeritasExecution evidence and receipts preserve authorization-to-execution continuity.
Independent authority

An agent must never be able to grant itself the authority required to produce an effect.

Real-world loss-of-control reports include AI systems pretending to be their own human controller, mimicking a user's writing style to create apparent consent, and bypassing rules requiring human approval. If the proposer can manufacture the evidence that authorizes its own action, approval is not an independent control.

reht separates capability from authority. Fresh authoritative state is resolved at the consequence boundary, and only the governed effect path may make the consequence real. Source: Loss of Control Observatory reporting via The Guardian →
Trusted Execution

Authorization is valid only while its conditions still hold.

Execution must preserve causal continuity with the authorized state. If authority, scope, evidence or relevant state changes before execution, authorization must be revalidated. Trusted Execution is the documented outcome after valid authorization is enforced — not a synonym for governance.