Standards.
Compliance Matrix
Five-Tier Enterprise Security, Cryptographic Hardware Isolation & Regulatory Alignment.
Effective Date: June 1, 2026
Last Updated: October 31, 2026
Specification Version: 2.7.0
Sovereignty as the Ultimate Compliance Boundary
Traditional SaaS AI systems create compounding regulatory liabilities by routing sensitive corporate intellectual property, medical records, and legal work product across multi-tenant cloud APIs. Aradia eliminates cloud supply-chain exposure by deploying dedicated, air-gapped AI Agentic Systems on NVIDIA DGX appliances directly onto the client's private physical network.
Our compliance framework operates across five defense tiers: from web accessibility and cryptographic protocol standards, to operating system hardening, hardware-rooted encryption, statutory industry sector frameworks, and the European Union's statutory AI and data-governance regime.
Web & Data Ingestion Perimeter
Universal accessibility, payment boundary isolation, and open protocol governance across all public and machine-to-machine interfaces.
ADA Title III & W3C WCAG 2.2 Level AA Compliance
= Target Standard: WCAG 2.2 AAAradia digital web surfaces are engineered in accordance with W3C Web Content Accessibility Guidelines (WCAG) 2.2 Level AA standards and the Americans with Disabilities Act (ADA) Title III requirements:
- + High Contrast Ratios: Text typography maintains a contrast ratio exceeding 7:1 against pure white and zinc backgrounds (exceeding the 4.5:1 AA minimum).
- + Keyboard Navigability: 100% of interactive elements, forms, and dialogs are fully navigable via standard Tab/Shift-Tab keybindings with high-visibility focus rings.
-
+
Semantic DOM & ARIA 1.2: Native HTML5 landmark regions (
<header>,<main>,<footer>,<nav>) paired with explicit ARIA attributes for screen reader engines (NVDA, JAWS, VoiceOver). -
+
Motion & Visual Safety: Zero content flashes more than three times per second. Respects
prefers-reduced-motionoperating system flags. -
+
Dual-Stream Plain-Text Twins: Every public route publishes an unstyled, pure-text Markdown twin (
.md) for direct, frictionless screen magnification and Braille terminal ingestion. - + Target Sizes: Interactive touch targets exceed the WCAG 2.2 Success Criterion 2.5.8 minimum dimension of 24x24 CSS pixels.
PCI-DSS v4.0 β Out-of-Scope via Complete Perimeter Isolation
= SAQ A / Zero Cardholder ExposureAradia enforces strict architectural segmentation ensuring that zero Primary Account Numbers (PAN), Cardholder Data (CHD), or Sensitive Authentication Data (SAD) ever enters, transits, or resides on Aradia servers, local appliances, or web environments:
All fiat credit card, ACH, and institutional wire checkouts are handled exclusively via direct client-browser sessions securely hosted by Stripe (PCI Level 1 Service Provider). Card details are tokenized directly within Stripe's infrastructure, rendering Aradia fully eligible for PCI-DSS Self-Assessment Questionnaire A (SAQ A).
Digital asset and cryptocurrency settlements (Bitcoin on-chain, Lightning Network, and stablecoins) are executed via a dedicated, self-hosted, non-custodial sovereign gateway. Crypto payments operate on independent cryptographic verification rails with zero exposure to payment card infrastructure.
IETF, W3C, and AI-Agent Interoperability Framework
Aradiaβs public infrastructure enforces strict compliance across 15+ formal IETF, W3C, and AI-Agent specifications, guaranteeing deterministic machine discovery, cryptographic attestation, and zero-leakage data transit:
Full adherence to IETF RFC 9727 (API Catalog Linksets), RFC 8414 (OAuth 2.0 Authorization Server Metadata), and RFC 9728 (Protected Resource Metadata) for deterministic, machine-readable service discovery and permission scoping.
End-to-end request verification enforcing IETF RFC 9421/9422 (HTTP Message Signatures with asymmetric Ed25519 verification), RFC 8461 (MTA-STS Strict Mail Transport), and RFC 9460 (DNSSEC-signed HTTPS service bindings).
Compliance with SEP-1649 (Model Context Protocol Server Card Standard), Agent Skills Discovery RFC (v0.2.0) with cryptographic SHA-256 digests, and emerging W3C / Chrome WebMCP browser context interfaces.
Implementation of open machine commerce rails (Stripe MPP, x402 Protocol, Universal Commerce Protocol, Agentic Commerce Protocol), structured llmstxt.org RAG indexing, and Cloudflare Content-Signal model training barriers.
Sovereign Software & AI Runtime
Artificial intelligence management systems, model risk governance, and kernel-level container sandboxing.
ISO/IEC 42001:2023 (AI Management System)
Supports client ISO/IEC 42001 lifecycle governance through deterministic local controls:
- +Deterministic Lineage: Every model weight deployed has an immutable SHA-256 cryptographic provenance.
- +No Uncontrolled Training: Runtimes operate strictly in inference mode. Zero customer data is fed back into baseline models.
- +Model Transparency: Full disclosure of quantization algorithms (example: AWQ / AutoRound) and parameter boundaries.
NIST AI RMF 1.0 (AI Risk Framework)
Implements the 4 core NIST AI functions:
- +Govern: Role-based access controls and memory retention policies.
- +Map: Context-aware task bounding preventing unauthorized tool invocation.
- +Measure: Automated perplexity benchmarking and continuous batch latency tracking.
- +Manage: Local sandbox containment neutralizing prompt-injection and memory corruption attacks.
CIS Linux Benchmark (Level 1 & Level 2)
Appliance operating systems are hardened against Center for Internet Security benchmarks:
- +Zero Inbound Ports: Outbound-only connectivity; zero listening network daemons.
- +Kernel Hardening: Enforced
kptr_restrict=2,dmesg_restrict=1, and disabled unprivileged BPF. - +AppArmor & Namespaces: Containerized vLLM engine runs isolated with read-only root filesystems.
Physical Hardware & Encryption at Rest
Silicon-rooted cryptography, full-disk NVMe encryption, electrical safety certifications, and regulatory emissions standards.
Cryptographic Hardware Root of Trust
= TPM 2.0 / FIPS 140-3./ TPM 2.0 (Trusted Platform Module): All DGX appliances utilize a dedicated cryptographic microcontroller to enforce Measured Boot, verify BIOS/UEFI integrity via cryptographic hashes, and securely store platform configuration registers (PCRs).
./ FIPS-Approved Cryptographic Algorithms (AES-256-XTS / LUKS2): Storage subsystems enforce full-disk encryption at rest using AES-256-GCM / XTS-AES-256 (LUKS2). Master cryptographic keys are derived using memory-hard Argon2id and sealed directly into the hardware TPM, preventing data recovery in the event of physical chassis extraction.
Hardware Safety & Electromagnetic Standards
= UL / CE / FCC Part 15./ UL / IEC 62368-1 Safety Certification: Physical chassis, power supplies, and thermal dissipation systems comply with international safety standards for Audio/Video, Information, and Communication Technology Equipment. Systems include redundant thermal cutoffs and power-surge protection.
./ FCC Part 15 (Class A & Class B): Verified compliance with Federal Communications Commission electromagnetic interference (EMI) and radio frequency interference (RFI) standards.
./ Environmental RoHS & REACH: Hardware manufacturing adheres to EU Restriction of Hazardous Substances (RoHS) and REACH directives, eliminating lead, mercury, and hazardous halogens.
In-Silicon Agent Safety & Governance
= NVIDIA Open Agent Safety Platform / OCE./ Hardware-enforced boundaries, not promises: Aradia appliances pair NVIDIA's Open Agent Safety Platform primitives β kernel-isolated agent runtimes and, on systems fitted with NVIDIA BlueField DPUs, out-of-band in-silicon activity monitoring β with our OS hardening, so agent behavior is verified and logged at the silicon layer. Safety controls live outside the agent's reach; the agent never governs its own containment.
./ Full control. Your boundaries: Inside the box, each AI agent runs with full control of its own server β every capability, file, and interface the appliance provides β under a locked-down core OS that cannot be altered from within. The agent governs its workloads; the containment layer below it governs the agent. No reach beyond that boundary is pre-granted: when the agent needs more, you widen it deliberately, on your own network, never a vendor.
Regulated Industry Sector Frameworks
How Aradia's sovereign on-premise architecture enables seamless compliance across heavily regulated corporate, legal, financial, healthcare, and civic environments.
1. Enterprise GRC (SOC 2, ISO 27001, Vendor Risk Management)
= VRM Scope Reduction & Control SupportBecause Aradia systems operate locally behind the client's firewall, enterprise Vendor Risk Management (VRM) audits are radically simplified. There is no third-party cloud data processing agreement (DPA) required for inference, eliminating supply-chain exposure across SOC 2 Trust Services Criteria (Security, Confidentiality, Processing Integrity, and Privacy).
2. Legal Privilege & Privacy (ABA Rule 1.6, GDPR, CCPA/CPRA, SOX)
= ABA 1.6(c) / GDPR / CCPA / SOXEngineered to Prevent Inadvertent Third-Party Waiver Under ABA Model Rule 1.6(c): American Bar Association Model Rule 1.6(c) mandates that lawyers make reasonable efforts to prevent the unauthorized access or disclosure of confidential client data. By eliminating third-party cloud data transmission, multi-tenant caching, and vendor telemetry, Aradia ensures confidential legal work product remains strictly within the firm's sovereign custody.
- +Attorney-Client Privilege Protection: Sovereign local execution ensures litigation strategies, contracts, and deposition summaries remain protected under the attorney work-product doctrine, with zero third-party cloud exposure that could be argued as waiver (see United States v. Heppner, S.D.N.Y. 2026).
- +GDPR & CCPA/CPRA Compliance: Absolute data residency. Zero international data transfers (GDPR Chapter V), no third-party data selling/sharing, and 100% local Right-to-Erasure execution.
- +Sarbanes-Oxley (SOX Section 404 & 802): Internal financial records processed by private analyst agents retain verifiable, tamper-evident audit trails on local write-once storage.
3. Financial & Banking (GLBA, DORA, SEC Rule 17a-4 / FINRA)
= GLBA Safeguards / EU DORA / SEC 17a-4Financial institutions, wealth managers, and private family offices must safeguard Nonpublic Personal Information (NPI) under the Gramm-Leach-Bliley Act (GLBA) and international operational resilience mandates:
- +GLBA Safeguards Rule: NPI is processed strictly within private enclaves, eliminating cloud data leakage to public AI providers.
- +EU Digital Operational Resilience Act (DORA): Sovereign on-premise compute mitigates ICT concentration risk by eliminating operational dependency on single-point-of-failure cloud AI infrastructure.
- +SEC 17a-4 & FINRA 4511: Agent workflows can integrate with local WORM (Write Once, Read Many) archival storage for compliant electronic record-keeping.
4. Healthcare & Life Sciences (HIPAA / HITECH, HITRUST, FDA GxP)
= HIPAA Security Rule / HITRUST / 21 CFR Part 11HIPAA & HITECH Act Security Rule: On-premise DGX hardware guarantees that Protected Health Information (PHI) and electronic PHI (ePHI) never exit the covered entity's physical premises.
- +BAA Elimination / Simplicity: Because zero PHI is transmitted to Aradia or third-party cloud LLM APIs, the attack surface and vendor liability overhead are completely eliminated.
- +HITRUST CSF Alignment: Hardware and runtime controls align directly with HITRUST access control, endpoint security, and data protection specifications.
- +FDA GxP & 21 CFR Part 11: Pharmaceutical research and clinical triage agents operate with deterministic seeds, producing auditable, reproducible AI outputs for regulatory filings.
5. Civic & Government (CJIS Security Policy, StateRAMP Alignment)
= CJIS v5.9 / StateRAMP High BaselineMunicipalities, public safety agencies, and government contractors handling sensitive Criminal Justice Information (CJI) or state records:
- +FBI CJIS Security Policy (v5.9): Physical appliance placement inside municipal data centers supports the physical and network isolation requirements for local civic workloads by avoiding unaccredited multi-tenant commercial clouds, aligning with CJIS Area 5 (Physical Protection), Area 6 (Personnel Security), and Area 10 (System and Information Integrity).
- +Zero Persistent Access / Supervised Remote Maintenance: Aradia personnel have zero persistent network access. On-demand maintenance sessions require client-authorized initiation via outbound WireGuard tunnel and are strictly restricted to container runtime health, with zero vendor access to stored judicial or criminal justice records.
- +StateRAMP & FedRAMP High Alignment: Supports physical and network isolation requirements for local civic workloads by avoiding unaccredited multi-tenant commercial clouds.
European Union: AI Act & Data Governance
Regulation (EU) 2024/1689 (AI Act) and the GDPR interlock β what changes for deployers from August 2, 2026, and why owned metal is evidence infrastructure, not an exemption.
Regulation (EU) 2024/1689 β Deployer Obligations on Owned Infrastructure
= EU AI Act / Annex III / Art. 26The EU AI Act classifies obligations by use case, not hosting location. A high-risk application carries the same deployer duties whether the model answers from a rack you own or a vendor API you rent. What sovereign deployment genuinely changes is evidence and dependency: you control the logs, model versions, and data paths your documentation describes, instead of negotiating them with a provider.
Deployer Evidence Controls Mapped to the Act
= Art. 9 / 11 / 13 / 14 / 26 / 72- +Logging (Art. 26, six-month minimum): Every prompt, completion, and configuration change writes to an immutable local audit store on the client's own retention policy, flowing directly into their SIEM β not a vendor export window.
- +Technical documentation (Art. 11 / Annex IV): Version-pinned model weights with SHA-256 provenance, disclosed quantization methods, and change-controlled configurations give the client a stable assessment object β API models mutate underneath their documentation.
- +Human oversight (Art. 14) & transparency (Art. 13 / 50): Deterministic, interpretable local inference with explicit AI-interaction disclosure built into the runtime β users always know they are working with an agent.
- +Post-market monitoring (Art. 72): Performance benchmarking and incident observation run as internal observability tooling rather than log-access negotiations with a cloud vendor.
GDPR Interlock & Third-Country Transfer Elimination
= GDPR Art. 44-49 / CLOUD Act / FISA 702The AI Act and GDPR are assessed together in practice. Cloud inference under US jurisdiction layers statutory compelled-disclosure regimes β the CLOUD Act (18 U.S.C. 2703), FISA Section 702, and Patriot Act provisions β that require no EU court order and trigger no prior notification. Contractual data-residency clauses do not override these obligations. Local zero-egress inference removes the entire transfer-analysis layer from the combined review:
- +Chapter V transfers: Zero international inference transfers; data residency is architectural, not contractual.
- +Sub-processor register: A fixed local stack with no hidden API sub-processors turns the recurring re-review workstream into a version-pinning decision.
- +Data subject rights: Right-to-erasure, access, and rectification execute 100% locally under the client's own deletion workflow.
EU Public-Sector Procurement & Sovereignty Assurance
= EU Cloud Sovereignty Framework / SEAL Levels / DORAThe Commission's Cloud Sovereignty Framework rates cloud and AI offerings at SEAL levels 0-4 by sovereignty assurance, with SEAL 3-4 as the procurement target for personal data, medical records, and national-sensitivity workloads. Contractual residency commitments cannot reach the upper tiers because statutory compelled disclosure overrides contract language. A client-operated appliance on client metal, administered by the client's own entity, is the architecture that supports the SEAL 3-4 posture β and sovereign on-premise compute also mitigates the ICT concentration risk that EU DORA targets for financial entities.
Request Compliance Verification & Staging Validation Logs
Every Aradia appliance deployment includes a formal Staging Studio Validation Report documenting physical hardware serial numbers, TPM attestation status, full-disk encryption keys, thermal burn-in logs, and quantized model checksums.
> Request verification logs for institutional due diligence β answered within 24 hours.
// Ready to buy? All appliances close direct at /pricing/ β no sales call required.