Pentagon AI Cybersecurity Framework Is Due to Congress Today — Here's What Defense Contractors Need to Know

By Hector Herrera | June 15, 2026 | Government

The US Department of Defense faces a congressional reporting deadline today, June 16, 2026, on a new AI cybersecurity governance framework mandated by the FY2026 National Defense Authorization Act (NDAA). The framework will impose structured security tiers on every AI and machine learning system the Pentagon acquires — and the compliance obligations will flow directly to the defence contractors, software vendors, and AI companies that supply those systems.

This is the most significant federal AI security requirement to emerge since the Cybersecurity Maturity Model Certification (CMMC) program was created for traditional defence software. The difference is that AI systems are now explicitly in scope.

What the NDAA Requires

According to an analysis by Crowell & Moring, the FY2026 NDAA directs the DoD to build a cybersecurity and physical security governance framework covering all AI and machine learning systems it acquires. The law sets a June 16 deadline for the DoD to report to Congress on the framework's status.

The framework mirrors CMMC — the Cybersecurity Maturity Model Certification — which established tiered security requirements for defence contractors handling controlled unclassified information. Under CMMC, contractors must achieve and certify specific security levels before winning DoD contracts. The new AI framework applies the same model to AI-specific risks: model integrity, training data security, adversarial robustness, and physical security of AI-enabled hardware systems.

Specifically, the provision targets:

• Generative AI systems embedded in defence contractor products or internal workflows
• Autonomous systems — drones, vehicles, logistics robots, and other AI-directed physical platforms
• Decision-support AI — systems that inform targeting, intelligence analysis, or logistics recommendations

Any vendor supplying these categories to the Pentagon now faces potential certification requirements before their systems can be procured.

Why This Is Significant

The DoD spends roughly $800 billion annually and is the world's single largest procurement customer for advanced technology. When it creates a compliance framework, the entire defence industrial base reorganizes around it. CMMC took years to implement but fundamentally reshaped how thousands of defence contractors handle information security. The AI extension is likely to do the same — with the added complexity that AI systems present attack surfaces CMMC was never designed to address.

The specific AI threats the framework is designed to counter include:

• Model poisoning — adversarial manipulation of training data that causes AI systems to behave incorrectly under specific conditions
• Adversarial examples — inputs crafted to fool AI perception systems (particularly relevant for autonomous vehicles and drone guidance)
• Data exfiltration through AI interfaces — using AI models as a vector to extract sensitive information from protected systems
• Supply chain attacks on AI components — compromising AI chips, firmware, or pre-trained models before delivery

These are not hypothetical. Nation-state adversaries — particularly China and Russia — have demonstrated interest in AI supply chain attacks as a complement to traditional cyber operations.

What Defense AI Vendors Should Expect

The framework report due today is a status update, not the final rule. Vendors should expect the actual compliance requirements to emerge over the next 12-18 months through a combination of DoD policy memos, Federal Acquisition Regulation (FAR) updates, and Defense Federal Acquisition Regulation Supplement (DFARS) clauses.

Practically, companies supplying AI to the DoD should prepare now:

• Document AI system provenance — where training data came from, which pre-trained models were used, which third-party components are embedded
• Implement model versioning and integrity verification — be able to prove that the AI system delivered is the one that was tested and certified
• Establish physical security controls for AI hardware in classified or sensitive environments
• Map existing CMMC controls to AI-specific risks — many controls will transfer; others will need AI-specific augmentation

Small and mid-sized defence contractors who have not yet engaged with AI-specific security requirements will face the steepest learning curve. Larger primes (Lockheed, Raytheon, Northrop, General Dynamics) have compliance teams actively tracking this and are likely already engaged in shaping the framework through the DoD's rulemaking process.

What to Watch

The congressional report due today will establish the official baseline. Watch for whether the DoD signals a fast-track implementation timeline — driven by urgency around adversarial AI threats — or a more deliberate multi-year rollout similar to CMMC's prolonged implementation. Also watch for whether the framework creates a separate certification body for AI security, or folds AI requirements into the existing CMMC third-party assessment infrastructure. That decision alone will determine how quickly requirements can practically be enforced.

Hector Herrera covers AI in government and defence policy for NexChron.