GOVERNANCE SPECIFICATION • REGULATORY ASSURANCE

Enterprise Compliance & Regulatory Framework Mapping

Deterministic AI Governance by Design

CAIRN ETP translates complex regulatory directives into cryptographic, runtime-enforced boundary checks. By placing deterministic execution controls in front of probabilistic model output, CAIRN produces the technical evidence an enterprise compliance review asks for.
1

EU Artificial Intelligence Act (EU AI Act)

CAIRN ETP produces technical and record-keeping evidence relevant to the obligations placed on High-Risk AI Systems (Articles 9–15). The mapping below is a starting point for a conversation with your assessor — compliance is a property of your organisation, established by an auditor, not of a tool.

EU AI Act Mandate Regulatory Requirement CAIRN ETP Implementation & Pillar Mapping
Article 9 Risk Management System
Continuous, iterative risk identification and mitigation across the entire AI agent lifecycle.
Pillar 01 (CAIRN)
Pillar 02 (PATH)

AST scan checks and default-deny gatekeepers evaluate and block high-risk execution vectors prior to model invocation.

Article 10 Data & Data Governance
Strict data provenance, authorized context boundaries, and protection against bias/poisoning.
Pillar 05 (ATLAS)

Versioned document grounding with deterministic vector retrieval ensuring strict context authorization.

Article 12 Record-Keeping & Traceability
Automatic, tamper-evident recording of events, decisions, and system logs throughout system operation.
Pillar 04 (COMPASS)

SHA-256 hash-chained decision records with signed checkpoints, tracking inputs, context hashes and the decision reached.

Article 14 Human Oversight
Ability for human operators to intervene, override, or halt automated AI agent execution safely.
Pillar 03 (BEACON)

Real-time telemetry broadcasting coupled with configurable human-in-the-loop (HITL) authorization gates.

Article 15 Cybersecurity & Accuracy
Resilience against unauthorized access, prompt injection, data exfiltration, and unexpected agent drift.
Pillar 07 (CRUCIBLE)

Tool execution is contained in a Docker container or a Windows Sandbox micro-VM with no network (--network none). Where neither runtime is present the isolation is absent, and CAIRN records the result as static-only rather than reporting a pass.

2

NIST AI Risk Management Framework (NIST AI RMF 1.0)

CAIRN ETP maps directly across the four core functions of the NIST AI RMF, enabling organizations to manage risks associated with agentic AI systems.

GOVERN Icon GOVERN GV-1.1 to GV-4.3
Pillars 01 & 02

Establishes organizational AI governance processes, transparent risk tolerance boundaries, and default-deny execution policies.

  • Policy Enforcement (CAIRN): Default-deny security gatekeepers evaluate AST syntax and block unauthorized agent actions before model invocation.
  • Capability Scoping (PATH): Intent decomposition and bounding turn budgets enforce strict enterprise operational perimeters.
MAP Icon MAP MP-1.1 to MP-2.3
Pillar 05 (ATLAS)

Contextualizes organizational risk by mapping data lineage, model dependencies, and institutional knowledge inputs.

  • Cryptographic Context Lineage: 4-layer ATLAS memory fabric maintains complete provenance over retrieved enterprise data chunks.
  • Source Attribution: Records which institutional document fragments were supplied to the model for a given decision, and hashes them into the decision record.
MEASURE Icon MEASURE ME-1.1 to ME-3.1
Pillars 03 & 04

Continuously quantifies, monitors, and evaluates AI system behavior, tool accuracy, and compliance posture.

  • Real-Time Telemetry (BEACON): Structured JSONL recording decision latency, assurance grades and human-in-the-loop triggers, in a form a SIEM can ingest.
  • Provable Audit Trails (COMPASS): Signed SHA-256 decision lineage hashes (cairn trace) enable independent verification by internal risk teams.
MANAGE Icon MANAGE MN-1.1 to MN-3.3
Pillars 06 & 07

Allocates risk mitigation resources, enforces deterministic runtime boundaries, and isolates tool actions.

  • Synthetic Verification (FORGE): Validates proposed model code generation and execution plans against deterministic unit checks.
  • Contained Execution (CRUCIBLE): Runs all autonomous agent tools inside resource-limited, air-gapped Docker and Windows Sandbox VMs.
3

ISO/IEC 42001:2023 (Artificial Intelligence Management System)

CAIRN ETP provides infrastructure that supports an organisation implementing and maintaining an ISO/IEC 42001 AI management system. The certification is the organisation’s, not CAIRN’s.

Control Clause Framework Focus How CAIRN ETP Enforces Compliance
Annex A.5 AI Impact Assessment
Systematic assessment of consequences for automated decision-making.		 Deterministic AST Validation: Scores execution risk vectors prior to execution, generating audit evidence for impact reporting.
Annex A.6 AI System Lifecycle
Governance across development, deployment, and operational runtime stages.		 Policy-as-Code Engine: Standardized policy definitions enforced uniformly across local workstations, enterprise clusters, and sovereign nodes.
Annex A.8 Data for AI Systems
Assurance regarding validity, quality, and authorization of RAG context.		 Evidence Fabric: Records which retrieved context informed a decision, so an answer can be traced back to the documents behind it.
Annex A.9 Traceability & Transparency
System auditability and explainability for external regulators.		 Cryptographic Event Ledgers: Hash-chained audit trails with signed checkpoints, exportable as evidence a third party can verify independently.
4

Financial Services & Enterprise Security Alignments

In addition to dedicated AI frameworks, CAIRN ETP reinforces existing enterprise security and banking standards.

SOC 2 Type II Evidence

Produces exportable cryptographic evidence bundles (audit.export) that an auditor can verify without running CAIRN, supporting your own evidence for the Security, Availability and Confidentiality criteria. CAIRN itself holds no SOC 2 attestation.

DORA: Operational Resilience

Digital Operational Resilience Act (Financial Sector): Provides operational independence from third-party cloud vendor outages: entitlement is verified offline, there is no phone-home and no remote kill-switch, and the governance layer runs on your own infrastructure. Where you configure a cloud model provider, that provider receives what you route to it.

FFIEC & PRA Guidelines

Supports model risk management review (e.g. OCC 2011-12 / SR 11-7) by placing a deterministic, policy-bound governance decision in front of every action an agent takes, and recording what it decided and why. The model remains probabilistic; the control around it does not.