AI and the False Claims Act: When Government Contractors Use Unvalidated AI

The federal government spent over $6.8 billion resolving False Claims Act cases in fiscal year 2025 — the largest annual FCA recovery in history. Of that total, cybersecurity-related fraud settlements accounted for $52 million across nine cases, a figure that more than tripled year-over-year for the second consecutive year. Whistleblowers filed 1,297 qui tam lawsuits in fiscal year 2025, the highest single-year total ever recorded.

Against that backdrop, a new category of FCA exposure is developing that should concern every government contractor using AI in contract performance: what happens when a contractor certifies compliance with contract requirements while deploying AI systems that have not been adequately validated?

This post examines how FCA liability attaches in the AI context, what “knowingly” means when the system making the error is autonomous, how the DOJ’s FOCUS Initiative (Fraud Oversight through Careful Use of Statistics) creates new whistleblower infrastructure targeting data-related fraud, and what contractors can do to manage the risk.

The False Claims Act: A Quick Primer

The False Claims Act, 31 U.S.C. §§ 3729–3733, imposes civil liability on any person who “knowingly presents, or causes to be presented, a false or fraudulent claim for payment or approval” to the federal government. The statute’s civil penalties are severe: each false claim triggers a penalty of $13,946 to $27,894 per claim (amounts are periodically adjusted for inflation), plus treble damages — meaning the government recovers three times its actual losses.

The FCA is not just a government enforcement tool. Private individuals — called relators — can file qui tam suits on the government’s behalf and collect a share of the recovery, typically between 15% and 30% of the government’s recovery if the DOJ intervenes, and up to 30% if the relator pursues the case alone. Whistleblowers received $4.5 million from cybersecurity-related FCA settlements in fiscal year 2025, up 68% from the prior year.

The statute covers three categories of conduct:

1. Knowingly submitting false claims for payment
2. Knowingly making false statements material to a false claim
3. Conspiring to defraud the government through false claims

The word “knowingly” does significant work in each of these. Understanding what it means — especially in the AI context — is essential to understanding where contractors are exposed.

What “Knowingly” Means — and Why AI Doesn’t Let You Off the Hook

One of the most common misunderstandings among contractors is that FCA liability requires intentional fraud. It does not. The statute defines “knowingly” to include three alternative mental states:

• Actual knowledge — the contractor knew the claim or statement was false
• Deliberate ignorance — the contractor deliberately avoided learning facts that would have revealed the falsity
• Reckless disregard — the contractor acted with reckless disregard for the truth or falsity of the information

That third prong — reckless disregard — is where AI creates special exposure. Reckless disregard means the contractor had reason to know that something might be wrong, failed to investigate, and submitted claims anyway. It does not require proof that anyone in the organization knew the AI was producing inaccurate or non-compliant outputs.

Consider a contractor using an AI system to generate compliance certifications, billing determinations, or performance reports submitted to the government. If that AI system has not been validated for accuracy and the contractor had reason to know it might produce errors — because, for example, the system was deployed without testing against the specific contract requirements, or because employees had flagged concerns about output quality — submitting AI-generated certifications without checking them could constitute reckless disregard of their accuracy.

The autonomy of AI systems complicates but does not resolve this analysis in the contractor’s favor. A contractor cannot avoid FCA liability by arguing that it was the AI, not the contractor, that made the false claim. The FCA’s “causes to be presented” language explicitly covers indirect false claims — meaning any party that causes a false claim to be submitted, even without submitting it directly, faces liability. A contractor that deploys an AI system knowing (or recklessly disregarding) that it may produce inaccurate claims has caused those false claims to be submitted.

The AI-Specific Compliance Risks

Several AI use cases create specific FCA exposure patterns that contractors should understand:

Cybersecurity Compliance Certifications

The most active area of FCA enforcement involving technology right now is cybersecurity. Under frameworks including CMMC (Cybersecurity Maturity Model Certification), DFARS 252.204-7012, and NIST SP 800-171, defense contractors must certify compliance with specific cybersecurity standards as a condition of receiving payment. When an AI system accesses, processes, or stores Controlled Unclassified Information (CUI) or Federal Contract Information (FCI) without meeting applicable security requirements, and the contractor has certified compliance, that certification is potentially false.

If the contractor deployed the AI system without assessing whether it met applicable cybersecurity requirements, without validating that the system’s data handling practices were compliant, or without disclosing the AI system to the government as required under its cybersecurity incident reporting obligations, the certification is potentially a knowingly false statement under the FCA.

AI-Generated Performance Reports and Billing

Contractors increasingly use AI to generate deliverables, draft reports, or calculate billing amounts under cost-plus or time-and-materials contracts. If an AI system inflates hours, misclassifies labor categories, or generates reports that misrepresent actual performance, and the contractor submits those AI-generated outputs without meaningful human review, FCA exposure follows from the submission of the inaccurate claims.

The deliberate ignorance prong is particularly relevant here. If a contractor implements an AI workflow specifically to reduce human review time and cost — which is often the point — and that reduced oversight results in inaccurate claims being submitted without correction, a court or jury may find that the contractor deliberately ignored warning signs that the system was producing errors.

AI in Healthcare and Benefits Contracting

The FCA’s heaviest enforcement activity has historically concentrated in healthcare. Government-contracted healthcare and benefits administrators that use AI for medical necessity determinations, billing code selection, or claims adjudication face layered FCA exposure: if the AI produces incorrect medical necessity determinations or billing classifications, and the contractor submits those determinations for payment without validating them, FCA liability can follow both from the direct false claims and from violations of underlying substantive regulations (Medicare coverage rules, TRICARE requirements, etc.) that are incorporated by reference into the contract.

In 2025, a military health benefits contractor and its parent paid $11.2 million to resolve FCA allegations that they falsely certified compliance with cybersecurity requirements in TRICARE contracts. AI-related billing and compliance issues are a natural extension of this enforcement pattern.

AI-Inflated Performance Claims

In competitive procurement contexts, contractors may use AI tools to analyze solicitations, generate technical proposals, or produce capability statements. If an AI tool generates performance claims — accuracy rates, error rates, processing speeds, coverage statistics — that are inaccurate, and the contractor submits those claims as part of a cost proposal or technical proposal without validating them, that submission may be a false statement material to the government’s contract award decision. Post-award FCA liability can flow from pre-award false statements if those statements influenced the government’s decision to award the contract.

The DOJ FOCUS Initiative: Data-Driven Fraud Detection

The Department of Justice’s Civil Division launched the FOCUS Initiative — Fraud Oversight through Careful Use of Statistics — to formalize its relationship with data miners who use sophisticated analytics to identify potential fraud patterns in government contracting data.

The initiative represents a significant development in FCA enforcement infrastructure. Historically, qui tam suits arose from insiders — employees or former employees who had direct knowledge of fraudulent practices. The FOCUS Initiative broadens the relator base to include professional data analysts who mine publicly available government contracting, billing, and payment data for statistical anomalies that correlate with potential fraud.

Under FOCUS, data miners can request meetings with the Civil Fraud Section to present their analytical capabilities and demonstrate how their data signals reliably correlate to fraud. The DOJ will prioritize working with data miners that demonstrate “insightful application of sophisticated technological capabilities to regulatory frameworks.”

For AI-using contractors, the FOCUS Initiative creates a specific new risk: the same AI and data analytics tools the government is adopting to detect fraud can be pointed at government contracting databases to surface anomalies in AI-generated billing, performance data, or certifications. Statistical patterns that might not be visible to a human reviewer — systematic biases in AI-generated billing calculations, improbably uniform error rates in AI-generated compliance certifications, unusual distributions in AI-assisted cost allocations — are exactly the kind of signals that data miners are being invited to bring to DOJ’s attention.

Practically, this means that even contractors who have not experienced internal whistleblowing should understand that external data analysis is now a meaningful source of FCA exposure. The FOCUS Initiative converts publicly available contracting data into a potential source of qui tam complaints.

The Materiality Standard

FCA liability requires not just a false statement but a material one. The Supreme Court’s 2016 decision in Universal Health Services, Inc. v. United States ex rel. Escobar established that a false statement is material if the government would have refused payment had it known the truth, or if it has a tendency to influence the government’s decision to pay.

In the AI compliance context, the materiality standard is increasingly being established by regulatory and contractual language. The Office of Management and Budget has declared compliance with AI safeguarding requirements “material to contract eligibility and payment” in guidance applicable to certain AI procurement contexts. That language is designed to pre-establish materiality as a legal matter — meaning contractors cannot argue that the government wouldn’t have cared about the AI system’s compliance status.

Similarly, cybersecurity compliance certifications under CMMC and DFARS have been treated as material by courts and by DOJ in FCA enforcement, because the government’s decision to award and pay on defense contracts is expressly conditioned on contractor compliance with those frameworks. AI systems that handle CUI or FCI without meeting applicable cybersecurity standards undermine the government’s fundamental condition of payment.

Whistleblower Exposure

The qui tam provisions of the FCA mean that current and former employees, IT staff, compliance personnel, and subcontractors are all potential relators. In the AI context, the most likely whistleblowers are people who:

• Implemented or monitored the AI system and know it was not adequately validated
• Raised concerns internally about AI output accuracy or compliance and were ignored
• Discovered that AI-generated certifications or performance reports were submitted to the government without human review
• Observed that an AI system was accessing or processing government data in ways that violated applicable cybersecurity requirements

The 1,297 qui tam filings in fiscal year 2025 reflect a relator bar that is increasingly sophisticated about identifying and pleading FCA cases. Specialized whistleblower attorneys actively recruit potential relators in technology contracting contexts. Employees who work on AI implementation in government contracting environments should be understood as potential whistleblowers — not because they are adversarial, but because the legal framework gives them a significant financial incentive to report compliance failures they observe.

Contractor AI governance programs should specifically address insider reporting: creating internal reporting channels, training staff on compliance obligations, and taking internal reports seriously are all practices that reduce both the likelihood of an undetected compliance failure and the likelihood that a concerned employee will go directly to a relator attorney rather than through internal channels first.

What Contractors Should Do Now

Conduct an AI audit before someone else does. Identify every AI system your organization uses in government contract performance. For each system, assess: Has it been validated for the specific use case? Does it handle government data? What certifications, if any, does its use implicate? This audit should be documented and treated as a compliance record.

Do not deploy AI that touches government data without a security assessment. AI systems that access, process, store, or transmit CUI or FCI are subject to the same cybersecurity requirements as any other IT system in the contract environment. The fact that the system is commercial off-the-shelf software or a third-party AI service does not exempt it. Get a written assessment from your cybersecurity team or an outside assessor before deployment.

Build human review checkpoints for AI-generated compliance certifications. Certifications submitted to the government should always have a named human reviewer who has actually reviewed the underlying data, not just the AI’s output. The record of that review should be maintained. This is the single most important practice for reducing FCA exposure from AI-generated submissions.

Review your cybersecurity compliance posture specifically for AI systems. CMMC, DFARS 252.204-7012, and NIST SP 800-171 apply to the entire contract environment, including AI systems. If your AI system uses APIs to access external services, stores data in cloud environments, or logs sensitive information, each of those elements needs to be assessed for compliance.

Train staff on AI-related FCA risk. Most FCA cases involving technology begin with someone inside the organization who knew there was a problem and either reported it or didn’t. Training compliance and program staff on what FCA exposure looks like in the AI context — and what internal reporting channels exist — is both a compliance measure and a risk mitigation strategy.

Take internal concerns seriously. If an employee raises concerns about an AI system’s accuracy, compliance, or data handling, that concern needs to be documented, investigated, and resolved. Ignoring an internal concern and continuing to submit claims is the clearest path to establishing the deliberate ignorance prong of FCA liability.

The Enforcement Trajectory Is Clear

The DOJ’s Civil Cyber-Fraud Initiative has expanded every year since its 2021 launch. The FOCUS Initiative broadens the relator base to include data miners who can identify fraud from external analysis. AI systems create new categories of inaccuracy and compliance risk that did not exist when many government contractors designed their compliance programs.

The combination of these trends — aggressive enforcement, sophisticated whistleblower support infrastructure, AI-specific compliance risks, and clear statutory liability for reckless disregard of compliance — means that government contractors using AI in contract performance face a materially different FCA risk environment than they did even two years ago.

The contractors that navigate this environment successfully will be those that treat AI governance as a compliance function, not just an operational one.

This post is for general informational purposes only and does not constitute legal advice. Reading this post does not create an attorney-client relationship. If you have questions about your specific situation, consult a qualified attorney.