Autonomous Labor / AI Tax Authorization
Evaluates whether an AI labor unit, autonomous action, registration, shutdown, or operating request has authority to proceed.
Open AI Tax DemoGovernment Evaluation SubmissionAuthority Before Action
An-Dub Governance Kernel
A deterministic governance-control kernel for evaluating whether AI outputs, agent instructions, or autonomous actions have authority before they proceed.
The model generates. An-Dub governs. The UI presents.
Control Layer
What An-Dub Is
An-Dub is a deterministic governance kernel for AI outputs.
It is not a chatbot, model, prompt wrapper, or moderation filter. An-Dub sits after generation and before release. The model may draft an answer, but An-Dub decides whether that output has authority to proceed.
The kernel is intentionally compact, fail-safe, and model-agnostic. It evaluates candidate outputs against active Policy/Rule Packs, Evidence Packs, and immutable governance boundaries. When authority is missing, evidence is absent, or a hard boundary fails, An-Dub does not guess. It blocks, routes, or marks the output as observation-only.
Deterministic decision behaviorFail-safe release posturePolicy/Rule Pack compatibleEvidence-aware enforcementISO-style control and audit compatibilityCompact kernel footprintModel-agnostic architectureStructured decision audit recordsGoverned decision states
The model generates. An-Dub governs. The UI presents.
Other AI systems play the game. An-Dub referees.
Release Authority
Why Normal AI Guardrails Are Not Enough
Most AI guardrails are designed to reduce bad behavior. They can block unsafe content, steer tone, apply moderation rules, or help the model avoid certain categories of response.
That is useful, but it is not the same as governed release authority.
Normal guardrails usually ask whether an output looks safe, allowed, or policy-compliant. An-Dub asks a stricter question: does this specific output have authority to be released under the active Policy/Rule Pack, Evidence Pack, and governance boundary?
That distinction matters.
A model should not approve itself. A prompt should not become policy. A UI should not decide release. A literal workaround should not pass just because it appears to satisfy the words of a rule.
An-Dub exists because AI systems are moving from conversation to action. In that environment, guardrails are not enough. Governments and controlled operators need a separate authority layer that can decide whether an AI output, agent instruction, or autonomous action is valid, invalid, or observation-only before it proceeds.
Normal guardrails influence generation.
An-Dub governs release.
Pipeline
How An-Dub Works
An-Dub separates generation from authority.
A model may create a draft, response, instruction, or proposed action. That output does not become valid simply because the model produced it. Before release, the output is passed through An-Dub's governance pipeline.
1
The model generates a candidate output.
2
The resolver identifies the active Policy/Rule Pack, Evidence Pack, scope, and evaluation context.
3
An-Dub evaluates the candidate output against the active rules, evidence requirements, and immutable governance boundaries.
4
The kernel returns a governed decision: VALID, INVALID, or VALID_OBSERVATION.
5
The UI presents the decision without secretly changing the result.
The model generates. An-Dub governs. The UI presents.
An-Dub does not ask whether the output sounds good. It asks whether the output has authority to proceed.
Controlled Evaluation
Evaluation Demo Surfaces
An-Dub is available for controlled evaluation through multiple MVP and POC demonstration surfaces.
The primary government evaluation use case is AI tax / autonomous labor authorization. This demo shows how a proposed AI labor-unit action can be evaluated against active policy, evidence, and authority requirements before it is allowed to proceed.
General Governed Chat
Allows evaluators to test user-supplied Policy/Rule Packs, Evidence Packs, and candidate outputs through the governed decision pipeline.
Open Governed Chat DemoAWS Lambda Kernel Runtime
Demonstrates that the An-Dub governance kernel is deployed as a callable backend runtime, not merely as a UI simulation.
Verify RuntimePolicy/Rule Pack + Evidence Pack Evaluation
Shows how active rules and supporting evidence affect VALID, INVALID, and VALID_OBSERVATION outcomes.
Open Policy/Evidence DemoThese demonstrations are not SaaS products or public chatbot features. They are controlled evaluation surfaces for reviewing the An-Dub governance model.
Authority before action.
Policy before release.
Evidence before trust.
An-Dub before operational authorization.
Protected Build Path
Controlled Development Posture
An-Dub was developed as a private governance-control project, not as a public SaaS product or customer-growth startup.
That posture was intentional.
Protect. Prove. Transfer.
Protect the kernel, source materials, Policy/Rule Pack structure, Evidence Pack structure, and evaluation artifacts from uncontrolled public release.
Prove the governance behavior through controlled demonstrations, repeatable decision states, evidence-aware enforcement, and live runtime evaluation.
Transfer the governance-control asset to authorized government evaluators or acquirers capable of hardening, red-teaming, validating, scaling, and deploying it.
An-Dub was kept non-public because the kernel is the asset. The objective was never public adoption first. The objective was to preserve the control pattern, prove that it works, and make it available for serious government evaluation.
Evaluation Path
Government Evaluation Path
An-Dub is being presented as a government evaluation and acquisition asset, not as a SaaS product.
The intended evaluation path is U.S. Government first: authorized agency reviewers, DARPA performers, national-security technical performers, or other approved government-aligned evaluators.
The current objective is controlled evaluation by technical teams that can assess, red-team, harden, validate, acquire, or deploy the governance-control pattern.
An-Dub was built to address a government-scale control problem: how to determine whether AI outputs, agents, autonomous actions, or AI labor-unit operations have authority before they proceed.
The acquisition frame is control.
Not chatbot performance.Not SaaS adoption.Not model competition.Control before deployment.Authority before action.Governance before autonomous systems scale.Contact
Evaluation Contact
An-Dub is available for controlled government evaluation.
Evaluation access may include live MVP demonstrations, POC review, Policy/Rule Pack and Evidence Pack testing, AWS Lambda runtime verification, and limited technical discussion of the governance-control architecture.
Runtime access does not authorize unrestricted source access, private evidence disclosure, internal gate traces, deployment credentials, or uncontrolled redistribution of evaluation materials.
Authorized reviewers, agency personnel, DARPA performers, national-security technical performers, or approved government-aligned assessment teams may request evaluation access through the contact below.
evaluation@an-dub.comAn-Dub is being presented for serious review as a governance-control asset.
Authority before action.
Policy before release.
Evidence before trust.
An-Dub before operational authorization.