Defense AI. Governed inside the authorization boundary.
Defense contractors and government agencies face strict data handling obligations: ITAR technical data cannot reach unauthorized AI endpoints, CUI must stay within CMMC-compliant systems, and every AI interaction with controlled information requires a tamper-resistant audit record. Arbitex puts a compliance-grade governance layer in front of every AI call — inspecting, enforcing, and logging before any data reaches a model. The data plane deploys inside your network. Nothing leaves your authorization boundary.
Capabilities
AI governance built for defense contractors and government environments.
Export-Controlled Data at the AI Boundary
Defense articles, munitions list items, and export-controlled technical data under the United States Munitions List must never reach unauthorized AI endpoints. Arbitex inspects every AI prompt in-path before a model processes it, and organizations author their own detection rules for the controlled identifiers in scope for their programs — USML category formats, program code words, and classification markings differ by contract and by customer, so they are defined by the contractor rather than shipped as fixed patterns. Analysts and engineers querying AI assistants about controlled defense technologies are inspected in-path. Violations are blocked or flagged before reaching any model endpoint — including commercial AI providers without the required authorization to handle controlled technical data.
CUI/CMMC 2.0 Enforcement
Defense contractors handling Controlled Unclassified Information must meet CMMC 2.0 requirements under DFARS 252.204-7012 to maintain DoD contract eligibility. Arbitex applies inspection and enforcement to every AI interaction at the gateway and records it in a tamper-evident audit entry. Arbitex does not ship a CMMC or CUI compliance bundle: the detection rules for your CUI categories are authored by your team, and the audit record gives your compliance staff a complete, tamper-evident evidence trail of what was sent, what was detected, and what was enforced — the raw material for DIBCAC assessments and prime contractor compliance audits.
NIST SP 800-171 Controls Mapping
NIST SP 800-171 defines 110 security requirements across 14 control families for protecting CUI in nonfederal systems and organizations. The controls Arbitex provides — access control over who can reach which models, a tamper-evident audit record of every AI transaction, and in-path enforcement — are the technical building blocks that several 800-171 families call for, including Access Control (3.1), Audit and Accountability (3.3), Identification and Authentication (3.5), and System and Communications Protection (3.13). Arbitex does not self-assert 800-171 coverage; your compliance team maps these controls to the requirement set and carries the audit evidence into system security plan artifacts and self-attestation submissions.
FedRAMP-Aligned Data Sovereignty
Federal agencies and defense contractors operating under FedRAMP authorization boundaries require that AI data processing occur within the authorized environment — not at a commercial AI provider's general-purpose infrastructure. Arbitex deploys the data plane inside your agency VPC, contractor network, or government cloud region: AI traffic inspection, DLP enforcement, and audit logging run entirely within your authorization boundary. The Arbitex control plane receives only telemetry and policy acknowledgments — no prompt content, no response content, no CUI ever transits Arbitex infrastructure. This architecture is designed for FedRAMP authorization boundary compliance at Moderate and High impact levels.
Air-Gap Outpost — SCIFs and Classified Networks
Sensitive Compartmented Information Facilities and classified networks operate in fully disconnected environments where any connection to commercial cloud infrastructure is prohibited. Arbitex Hybrid Outpost deploys the entire data plane inside your SCIF or classified network segment: AI governance — routing, DLP inspection, policy enforcement, and audit logging — runs with no persistent connection to the Arbitex control plane or any external endpoint. Policy bundles are cryptographically signed and delivered through your approved offline transfer procedures. The gateway operates fail-closed: if the policy bundle cannot be validated, all AI traffic is blocked until the condition is resolved. Approved AI model endpoints can be proxied locally or accessed via one-way data diode connections.
Zero-Trust Identity for Defense Programs
Defense programs require strict identity compartmentalization at every AI access point — analysts on one program must not be able to query AI systems with data from another. Arbitex enforces zero-trust identity at the gateway: every AI request is authenticated via SAML 2.0, OIDC, or CAC/PIV-backed certificate assertion before DLP inspection runs. Role-based access controls enforce program isolation — limiting which personnel can query which model endpoints and data categories. SCIM 2.0 provisioning keeps access grants synchronized with your enterprise directory as programs staff up, rotate, or wind down. WebAuthn/FIDO2 hardware token support satisfies phishing-resistant MFA requirements for high-assurance defense environments. Every identity assertion is logged alongside the AI transaction for the full audit trail.
How it works
Deploy inside your defense network or classified environment
The Arbitex data plane installs in your defense network using Docker Compose or Kubernetes — inside your contractor facility VPC, government cloud region, or classified network segment. For SCIF and air-gap environments, the Outpost package installs on locally provisioned infrastructure with no external network dependency. All AI traffic from analyst workstations, program offices, and automated workflows routes through the gateway before reaching any model endpoint. Inspection, enforcement, and audit logging run entirely inside your authorization boundary. No prompt content, response content, or controlled data transits Arbitex infrastructure.
Inspection and enforcement run in-path before any model processes it
Your configured compliance bundle activates the relevant control sets. Tier 1 matches structured identifiers — including any export-control or program-specific patterns your team has authored as custom org rules — with checksum or format validation where the identifier supports it. Tier 2 applies ML-based detection to identify personal and credential data in free-text analyst queries. Contextual validation at Tier 3 resolves ambiguous detections, distinguishing routine defense terminology from genuinely sensitive content. Enforcement actions — block, redact, or route-to-review — execute before any data reaches the model, and every decision is logged with its policy reference.
Audit evidence ready for CMMC assessment and DFARS compliance
The tamper-proof audit log accumulates a complete evidence trail for every AI interaction involving controlled or CUI-category data. Signed exports record each enforcement action with its policy reference and timestamp — an evidence trail your compliance team can carry into DIBCAC assessments, prime contractor audits, and DFARS 252.204-7012 compliance submissions. SIEM integrations (Splunk, Microsoft Sentinel, Elastic) deliver enforcement metrics for continuous monitoring of controlled data handling across defense program teams.
Six frameworks. One policy layer.
Each compliance obligation maps to a specific Arbitex capability. All bundles are active simultaneously — no separate configuration per framework, program, or impact level.
CMMC 2.0 governs CUI handling for the defense industrial base. Arbitex does not ship a CMMC bundle; it supplies AI access control, in-path enforcement, and tamper-evident audit logs that your compliance team uses as DIBCAC assessment evidence.
Defines 110 CUI security requirements across 14 control families. Arbitex supplies access control, enforcement, and tamper-evident audit evidence for AI interactions; mapping that evidence to specific requirements remains your compliance team’s responsibility.
Governs export-controlled defense articles and USML technical data. Arbitex does not ship ITAR detectors; contractors author org DLP rules for the controlled identifiers in scope, which then run through all three inspection tiers with full audit evidence.
Data plane deployment inside agency or contractor authorization boundary. Designed for FedRAMP Moderate and High impact level requirements. No prompt or response content transits Arbitex-controlled infrastructure outside the authorized environment.
Covered defense information handling controls and rapid reporting workflows. Audit exports formatted for DFARS incident reporting and cloud service provider review. CDI detection and enforcement active for all AI interactions in covered contractor systems.
Deployment architecture designed for DoD IL4 (CUI) and IL5 (National Security Systems) data handling requirements. Data plane operates within DoD authorization boundaries with identity assertion logging and fail-closed enforcement.
Related Resources
Ready to put governance in front of your defense program AI?
Talk to an Arbitex engineer about custom detection rules for export-controlled data, CMMC 2.0-aligned deployment, audit evidence for NIST SP 800-171 programs, and air-gap Outpost configuration for your SCIF, classified network, or defense contractor environment.