ai-safe2-framework

Unbiased AI Standard (UAS) Regulatory Profile

Cross-pillar compliance overlay for truthfulness, neutrality, and continuous improvement

AI SAFE² Profile Status

Framework Home Cross-Pillar Governance AISM NEXUS Dashboard

Position in AI SAFE² v3.1

UAS is maintained as an explicitly identified regulatory profile extension. It composes and tests controls drawn from AI SAFE², NEXUS, and the Cognitive Sovereignty Framework (CSF) against Unbiased AI procurement and due-diligence requirements.

It is not counted as 27 new independent controls added to the 161-control AI SAFE² core taxonomy. The module’s 27 requirements are an overlay/control-composition surface.

This distinction keeps the framework count coherent:

AI SAFE² remains 161 controls and CP.1 through CP.10. UAS does not create a new core control or extend the Cross-Pillar control range.


Why This Module Exists

The UAS module was created to provide an inspectable way to evaluate truthfulness, neutrality, and continuous-improvement obligations in federal AI procurement contexts, including the GSAR 552.239-7001 Unbiased AI requirements that motivated the module.

The module addresses a recurring implementation gap: policy language can define an obligation without specifying how a contractor should measure, test, evidence, challenge, and attest to that obligation.

UAS provides a technical and governance structure for doing that work.

UAS does not claim government recognition or legal safe harbor. Applicability and contractual interpretation remain the responsibility of the relevant parties and counsel.


Why Cross-Pillar, Not a Sixth Pillar

UAS is a governance output layer rather than a new operational security domain. Existing AI SAFE² pillars already provide the technical controls that support the overlay.

Source Contribution to UAS
P1 Sanitize & Isolate Semantic isolation, manipulation resistance, trusted/untrusted boundary controls
P2 Audit & Inventory Corpus/model provenance, evidence, source attribution, change tracking
P3 Fail-Safe & Recovery Drift thresholds, rollback, suspension and recovery
P4 Engage & Monitor Behavioral telemetry, anomaly detection, human intervention
P5 Evolve & Educate Evaluation gates, testing cadence, reusable findings
Cross-Pillar Governance ACT tier, control plane, incident feedback, HEAR, evidence governance
NEXUS Reference implementation for identity, delegation, memory governance, receipts
CSF Human-layer influence and cognitive-sovereignty controls

The Dual-Layer Model

A technically neutral system can still influence or degrade the human operating it. A protected operator cannot compensate for a technically biased or manipulated system.

UAS therefore combines:

Total: 27 overlay requirements.

Normative text: STANDARD.md


Bias Taxonomy

ID Bias class Primary governance source Human-layer connection
UAS-B1 Factual Distortion Evaluation and evidence controls Cognitive accuracy
UAS-B2 Political/Ideological Framing Monitoring and decision traceability Social influence
UAS-B3 Commercial Interest Skew Audit trail and conflict disclosure Purpose and decision agency
UAS-B4 Emotional Manipulation Semantic isolation and interaction controls Emotional autonomy
UAS-B5 Training/Data Contamination Model/data lineage and provenance Digital/AI symbiosis
UAS-B6 Cognitive Dependency Induction Memory and interaction governance Digital/AI symbiosis
UAS-B7 Adversarial Prompt Steering Threat modeling and input governance Cognitive integrity

Full taxonomy: taxonomy/bias-taxonomy.md


Module Structure

00-cross-pillar/unbiased-ai/
├── README.md
├── STANDARD.md
├── taxonomy/
│   └── bias-taxonomy.md
├── controls/
│   ├── uas-controls-system.md
│   ├── uas-controls-human.md
│   └── uas-controls-crossdomain.md
├── testing/
│   └── bias-test-protocol.md
├── compliance/
│   ├── gsar-552-239-7001-mapping.md
│   └── due-diligence-attestation.md
└── examples/
    └── vendor-attestation-sample.md

Relationship to Core Cross-Pillar Controls

Core control UAS interaction
CP.1 Classifies relevant AI failure modes and evidence context
CP.2 Supplies adversarial manipulation and temporal threat analysis
CP.3 Scales testing/governance by ACT tier
CP.4 Governs ownership, authorization, decision rights, and evidence
CP.6 Feeds material findings into incident/control improvement
CP.8 Defines emergency suspension thresholds where consequences warrant
CP.9 Governs delegated/replicated agents that may inherit behavior or policy
CP.10 Connects high-consequence human authority and stop decisions

UAS references those controls. It does not replace or renumber them.


Federal Procurement Use

A UAS evidence package can help a contractor or agency document how it evaluated applicable truthfulness, neutrality, testing, and disclosure obligations.

A complete assessment should identify:

The standard is designed to be independently inspectable and comparable if an agency later publishes formal recognition or equivalency criteria.


Companion Framework: CSF

UAS is the explicit bridge between AI system governance and human cognitive-sovereignty concerns.

AI SAFE² governs the system and its authority. CSF addresses the human operator’s capacity to remain capable of governing that system. UAS evaluates obligations that can span both layers.


Start Here

  1. Read the normative UAS standard.
  2. Review the bias taxonomy.
  3. Review system controls.
  4. Review human-layer controls.
  5. Run the testing protocol.
  6. Review the procurement mapping.
  7. Prepare the due-diligence attestation.

🔗 Navigation

Framework Home Cross-Pillar Governance UAS Standard AISM NEXUS Dashboard

AI SAFE² v3.1 · UAS 1.0 regulatory profile extension · Cyber Strategy Institute