Why AI for Defense Contractors Is Different
Defense contractors are among the most knowledge-intensive organizations on earth. A mid-sized prime contractor or Tier 1 sub manages thousands of documents across dozens of active programs: contracts, modifications, technical specifications, past performance records, compliance documentation, teaming agreements, subcontractor obligations, and export control records.
And yet, as of 2025, most of that knowledge exists in siloed file servers, email archives, and the minds of individual employees — completely inaccessible to AI.
The adoption of artificial intelligence in the defense industrial base is accelerating rapidly. The Department of Defense itself has established the Chief Digital and Artificial Intelligence Office (CDAO) and has published AI ethics principles, acquisition guidance, and an AI adoption roadmap. Defense contractors are under pressure from their government customers to modernize — and AI is increasingly part of that modernization conversation.
But deploying AI in a defense contractor environment is not the same as deploying it in a commercial enterprise. The stakes are fundamentally different. Data handling mistakes don't just result in regulatory fines — they can result in program loss, debarment, security investigations, and reputational damage that ends a company's ability to compete for government work.
IT leaders, program managers, COOs, and security officers at defense primes and Tier 1 subs who need to understand what AI can legitimately do in a defense environment, what compliance requires for CUI and sensitive data, which deployment models fit which security posture, how to evaluate vendors for defense-grade security, and where to start with a pilot that delivers measurable value quickly.
This is not a guide to AI in the abstract. It is a practical guide for defense contractor operations teams who need to deploy AI responsibly, compliantly, and effectively.
The Real Risks of Generic AI Tools in Defense Environments
Before addressing what AI can do for defense contractors, it is worth being direct about what the wrong AI tools can do to defense contractors.
Research consistently shows that 60–80% of enterprise employees use AI tools their IT department hasn't sanctioned. In defense environments, cleared personnel use personal ChatGPT accounts, free Copilot tiers, and consumer AI assistants — pasting contract modifications, draft technical approaches, and proprietary specifications into them.
Controlled Unclassified Information is a formal category defined under 32 CFR Part 2002. Contracts require contractors to protect CUI per NIST SP 800-171 and CMMC. Routing CUI through a commercial AI tool's external servers is a direct violation of these obligations — whether or not it feels like one in the moment.
Government contracts require demonstrable oversight of business processes. If AI influences a contracting decision or compliance determination, auditors will ask what AI was used, what data it accessed, and who approved the output. Generic AI tools produce no such record.
ITAR controls the export of defense articles and related technical data. If a commercial AI provider's infrastructure is operated by or accessible to foreign nationals, sending ITAR-controlled technical data to it may constitute an unlicensed export — a risk the DDTC is expected to address in formal guidance within 12–18 months.
Every one of these interactions sends data to a commercial AI provider's servers. That data may be used to train future models. It definitely leaves the contractor's controlled environment. And it may constitute a CUI violation, an ITAR export control violation, or a contract breach — none of which have trivial consequences.
The Regulatory Landscape: CMMC, CUI, ITAR, and FAR/DFARS
Defense contractors deploying AI must navigate five overlapping regulatory frameworks. Understanding these frameworks — and how they interact — is essential to building a defensible AI compliance posture.
| Framework | What It Covers | Key Requirement for AI |
|---|---|---|
| CMMC 2.0 | DoD certification model for protecting CUI | Level 2 requires all 110 NIST SP 800-171 practices |
| NIST SP 800-171 | Baseline security requirements for nonfederal systems | Access control, audit logging, encryption at rest/in transit |
| CUI Program (32 CFR 2002) | Standardized handling of sensitive government information | Access controls, marking, storage, transmission protections |
| ITAR (22 CFR 120–130) | Export of defense articles and technical data | No foreign-national access; no use in model training |
| FAR / DFARS | Federal contracting information-handling clauses | NIST 800-171 implementation; 72-hour incident reporting |
CMMC — Cybersecurity Maturity Model Certification
CMMC is the DoD's framework for ensuring defense contractors adequately protect CUI in their information systems. As of 2025, CMMC 2.0 is being phased into contract requirements, with Level 2 applicable to most defense contractors handling CUI. CMMC Level 2 requires compliance with all 110 practices from NIST SP 800-171. The most relevant practice domains for AI deployment:
- Access Control (AC): AI systems must enforce role-based access controls restricting data access to authorized personnel only.
- Audit and Accountability (AU): Every AI interaction — query, response, data accessed — must be logged in an immutable, reviewable audit record.
- Configuration Management (CM): AI platform configurations must be documented and controlled.
- Identification and Authentication (IA): AI access must be controlled through enterprise identity management.
- System and Communications Protection (SC): AI systems handling CUI must be isolated from unauthorized network access.
NIST SP 800-171
NIST SP 800-171, "Protecting Controlled Unclassified Information in Nonfederal Systems and Organizations," defines the baseline security requirements for contractors who handle CUI and is the foundation of CMMC Level 2. Key requirements for AI deployment include limiting system access to authorized users and devices (3.1.1), creating and retaining system audit logs (3.3.1), monitoring and protecting communications at external boundaries (3.13.1), and protecting the confidentiality of CUI at rest through encryption (3.13.16).
CUI Program — 32 CFR Part 2002
The National Archives and Records Administration (NARA) administers the CUI Program, standardizing how executive branch agencies and their contractors handle sensitive government information. Relevant CUI categories include Controlled Technical Information, Export Controlled information, Procurement Sensitive information, Privacy data, and Law Enforcement Sensitive information. Any AI system that accesses, stores, or processes CUI must be deployed consistent with 32 CFR Part 2002 handling requirements.
ITAR — International Traffic in Arms Regulations
ITAR (22 CFR Parts 120–130) controls the export and temporary import of defense articles and related technical data. For AI deployment, ITAR creates two specific risks: data residency — technical data must not be transferred to systems accessible by foreign nationals, which is an inherent risk with internationally distributed commercial AI infrastructure — and model training — if a provider uses customer data to train or improve models, ITAR-controlled data used in training could constitute an unauthorized export. Self-hosted deployment eliminates this risk entirely.
FAR and DFARS
The Federal Acquisition Regulation and its Defense Supplement contain clauses affecting how defense contractors must handle information. Relevant to AI: FAR 52.204-21 (Basic Safeguarding of Covered Contractor Information Systems), DFARS 252.204-7012 (Safeguarding Covered Defense Information and 72-hour cyber incident reporting), and DFARS 252.204-7019/7020 (NIST SP 800-171 DoD Assessment requirements). Any AI system handling covered defense information on contractor systems subject to these clauses must be compliant with them.
The Three Deployment Models for Defense Contractor AI
The fundamental question for any defense contractor evaluating AI is: where does the data go? The answer determines which deployment model is appropriate.
| Model | Best Suited For | Key Limitation |
|---|---|---|
| Private Cloud (GovCloud) | CUI without classified data; faster deployment, lower infrastructure overhead | Data resides in a cloud environment, not contractor hardware |
| On-Premise Self-Hosted | ITAR-controlled technical data; highly sensitive programs; strict data residency needs | Contractor must manage infrastructure (8-core CPU / 64GB RAM minimum, GPU optional) |
| Air-Gapped Network | Classified data, SAPs, SECRET/TOP SECRET facilities | Requires physical media transfer for updates; needs a dedicated implementation engagement |
The Six Highest-Value AI Use Cases for Defense Contractors
Based on deployment experience with defense contractor pilots, these are the six use cases that consistently deliver measurable ROI in the first 90 days, using Sphere's Company Brain platform.
Ready to Scope a 90-Day Defense AI Pilot?
Sphere's engineers deploy AI in cleared contractor environments — GovCloud, on-premise, or air-gapped. Talk to us under NDA about your architecture and compliance requirements.
Request a Security Briefing Contact Sphere →How to Evaluate an AI Platform for Defense Use
Not all AI vendors are equipped to serve defense contractors. When evaluating an AI platform for deployment in a defense environment, ask these questions.
Security and Deployment
- Does the vendor offer self-hosted or on-premise deployment, or only cloud-hosted SaaS?
- In self-hosted mode, does any data leave the contractor's environment for any reason (telemetry, logging, model updates)?
- Is air-gapped deployment available for classified or highly sensitive environments?
- Is the vendor able to deploy in AWS GovCloud or Azure Government?
- What network connectivity does the platform require in self-hosted mode?
Data Handling
- Does the vendor use customer data to train or improve their AI models? (Disqualifying for most defense programs.)
- Where is data stored at rest, and is it encrypted?
- Does the platform support data retention and deletion policies required by government contracts?
- Can the platform tag and enforce access controls based on CUI category?
Compliance and Governance
- Does the platform maintain immutable audit logs of all AI interactions?
- Are audit logs exportable for government oversight review?
- Does the vendor have a published NIST 800-171 System Security Plan (SSP)?
- Has the vendor undergone a CMMC Level 2 assessment, or can they support your assessment process?
- Does the vendor have cleared US-person technical staff who can support sensitive program deployments?
Access Control and Identity
- Does the platform integrate with enterprise identity management (Active Directory, LDAP, SAML)?
- Can access to specific documents or knowledge domains be restricted by user role or clearance level?
- Does the platform enforce need-to-know access controls at the document level?
Vendor Trustworthiness
- Is the vendor a US-owned company with US-person employees?
- Does the vendor have experience deploying in cleared contractor environments?
- Is the vendor willing to sign an NDA before technical briefings?
- Can the vendor provide references from defense contractor customers?
- Does the vendor have an active or in-progress FedRAMP authorization?
The CMMC Compliance Checklist for AI Deployment
Use this checklist when preparing an AI deployment for CMMC Level 2 assessment, mapped to the NIST SP 800-171 practice families most relevant to AI systems.
- AC.1.001 — Limit information system access to authorized users.AI platform integrated with enterprise identity management; access requires authentication.
- AC.1.002 — Limit access to authorized transactions and functions.Role-based access controls configured; each user accesses only authorized knowledge domains.
- AC.2.006 — Use non-privileged accounts for non-security functions.AI administrator access separated from standard user access.
- AU.2.041 — Ensure user actions can be traced to individual users.All AI queries and responses logged with user identity, timestamp, and data accessed.
- AU.2.042 — Create and retain system audit logs.Audit logs retained for a minimum period aligned to contract requirements, typically 3 years.
- AU.3.045 — Review and update logged events.Audit log review process established; anomalous activity flagged for security review.
- CM.2.061 — Establish and maintain baseline configurations.AI platform configuration documented and baselined; changes controlled through change management.
- CM.2.064 — Establish a security configuration checklist.AI platform security configuration checklist established and reviewed periodically.
- IA.1.076 — Identify and authenticate users before access.AI platform requires enterprise credentials (SSO/MFA) before access.
- IA.3.083 — Use multi-factor authentication.MFA enforced for all AI platform access when handling CUI.
- SC.1.175 — Protect communications at external boundaries.Self-hosted or GovCloud deployment ensures CUI does not cross uncontrolled network boundaries.
- SC.3.177 — Employ FIPS-validated cryptography.Data encrypted at rest and in transit using FIPS 140-2 validated encryption.
- SC.3.187 — Establish and manage cryptographic keys.Encryption key management documented and controlled.
- SI.1.210 — Identify, report, and correct system flaws.AI platform update and patching process established; self-hosted deployments receive security patches.
- SI.2.216 — Monitor systems to detect attacks and anomalies.AI interaction monitoring for anomalous access patterns; alerts configured for security team.
Common Objections from Defense IT Leaders
"We already have Microsoft Copilot through our Microsoft 365 subscription."
Standard Microsoft 365 Copilot routes data through Microsoft's commercial cloud infrastructure, which is not designed for CUI handling. Microsoft 365 GCC High is more appropriate for CUI, but even GCC High Copilot has no access to your proprietary knowledge repositories, cannot be self-hosted or air-gapped, and does not maintain the interaction audit log required for full CMMC compliance. Copilot answers questions about your Microsoft 365 content; Company Brain answers questions about your company's knowledge.
"Our security team will never approve an AI platform."
Security teams are right to be cautious. The choice is not between "AI" and "no AI" — employees are already using AI without approval, through personal accounts. The real choice is between unsanctioned AI with no controls and sanctioned AI with full security controls, audit trails, and data governance. Frame the conversation correctly and most security teams become advocates for a controlled solution.
"We don't have the IT resources to manage a self-hosted deployment."
Self-hosted AI platforms have become significantly easier to deploy than 18 months ago. Company Brain's self-hosted deployment requires standard Linux server administration skills, not a specialized AI infrastructure team. Ongoing maintenance involves periodic security patches and model updates — deliverable via physical media for air-gapped environments. Sphere provides dedicated implementation support and can train an IT team on platform administration in 2–3 days.
"What if the AI gives a wrong answer about a compliance requirement?"
The answer is design, not avoidance. Company Brain is a research and drafting tool, not an autonomous decision-maker. Every answer includes citations to the source documents that informed it. Users are trained to treat outputs as a starting point for human expert review. For high-stakes compliance questions, the AI's value is surfacing relevant information quickly so a human expert can make the final determination.
"We don't know where to start."
Start with a single, high-value, low-risk use case: typically proposal support or new-hire onboarding. These have clear, measurable outcomes, don't involve the most sensitive data classifications, and generate visible quick wins that build organizational confidence. A 90-day pilot with a focused use case is the right starting point for virtually every defense contractor.
Implementation Roadmap: From Pilot to Enterprise Deployment
Frequently Asked Questions
Does Sphere Inc. use our data to train AI models?
No. In self-hosted and on-premise deployments, your data never reaches Sphere Inc.'s systems at all. In cloud-hosted deployments, your data is processed only to provide your service and is never used for AI model training. This is a contractual commitment, not just a policy statement.
What AI models does Company Brain use?
Company Brain supports multiple underlying AI models, including GPT-4o, Claude, and open-source models that can be run entirely on-premise without external API calls. For air-gapped environments, Sphere supports self-hosted open-source models that require no external connectivity.
Is Company Brain FedRAMP authorized?
Sphere Inc. is pursuing FedRAMP authorization. In the interim, Company Brain's GovCloud deployment runs on FedRAMP-authorized cloud infrastructure (AWS GovCloud / Azure Government). Current FedRAMP status documentation and SSP are available on request under NDA.
How does Company Brain handle document updates? If a contract is modified, does the knowledge base automatically update?
Yes. Sphere Knowledge supports automated synchronization with connected document management systems. When a document is updated in SharePoint, Confluence, or a connected file system, the knowledge base updates automatically within a configurable refresh window, typically 15 minutes to 24 hours depending on configuration.
Can we restrict certain employees from accessing certain programs' information?
Yes. Sphere Knowledge supports granular, document-level access controls, configurable at the level of individual documents, document categories, program folders, or classification tags. A user without access to a specific program's documents will not receive information from those documents in any query response.
What happens to our data if we stop using Company Brain?
In self-hosted and on-premise deployments, your data never leaves your environment and remains entirely under your control at all times. In cloud-hosted deployments, Sphere provides a full data export in standard formats and certified deletion of all customer data upon contract termination.
Does Company Brain work on classified networks (SIPRNet)?
Air-gapped deployment is compatible with classified network environments. SIPR-specific deployment requires a dedicated implementation engagement — contact Sphere to discuss specific classified network requirements.
Conclusion and Next Steps
The question for defense contractors is no longer whether to deploy AI. Employees are already using it. Competitors are deploying it. The DoD is demanding it. The real question is whether AI is deployed responsibly — with appropriate security controls, compliance architecture, and governance — or whether the unsanctioned adoption already happening is allowed to continue unchecked.
The good news: responsible AI deployment for defense contractors is achievable, and the path is clearer than it has ever been. Self-hosted and air-gapped deployment models eliminate the core data exposure risk. CMMC-aligned audit and access control capabilities address the compliance challenge. Focused initial use cases — proposal support, knowledge continuity, compliance Q&A — deliver measurable ROI within 90 days.
1. Assess your shadow AI exposure. Ask IT how many employees have accessed ChatGPT, Copilot, or other AI tools from your network in the past 90 days — the number is almost always higher than leadership expects.
2. Identify your highest-value use case. Proposal teams? Contract compliance? New-hire onboarding? Start where the most time is lost to manual knowledge work.
3. Request a security briefing. Talk to Sphere under NDA — walk your security and IT teams through the architecture, get compliance questions answered, and determine the right deployment model.
4. Scope a 90-day pilot. One team, one use case, measurable outcomes defined before deployment begins.
Request a Security Briefing
US-based team · NDA available · Air-gapped demo on request. Talk to Sphere about deploying Company Brain in your defense environment.
Request a Security Briefing Contact Sphere →