by Henry Fina and Matthew P. Suzor
Left to right: Henry Fina and Matthew P. Suzor (photos courtesy of Miller Shah LLP)
The explosion of the Artificial Intelligence market has drawn capital investment from almost every corner of the economy. The federal government is no exception. Between FY 2022 and 2023, the potential value of federal AI contracts increased from approximately $356 million to $4.6 billion. In July 2025, the Trump Administration released its AI Action Plan, outlining government initiatives to aggressively deploy AI in the health and defense sectors. Accordingly, the Department of Health and Human Services (HHS) and Department of Defense (DoD) have increased funding allocations toward AI contracts. As contractors compete for increasingly valuable awards with limited oversight, the potential for misrepresented capabilities and compliance gaps grows. While the industry’s strong tailwinds may translate into lucrative opportunities for investors and entrepreneurs, for qui tam litigators, the expansion of publicly contracted AI services signals a new frontier for False Claims Act (FCA) enforcement. In turn, the FCA will be essential in ensuring accountability as federal agencies gradually adjust oversight mechanisms to handle the inconsistent reliability and limited technological opacity of AI models.
Most FCA cases levied against AI contractors would likely stem from false or fraudulent representations of a model’s capabilities. Misrepresentations may include inflated claims regarding the accuracy of a model’s outputs, concealment of bias or synthetic training, or insufficient data privacy and security standards. Whether an AI model is used for surveillance and intelligence by the DoD or for the Centers for Medicare and Medicaid Services (CMS) to review claims, there are concerns beyond the technical effectiveness of AI outputs.
There are deeper concerns about the accuracy, accountability, and integrity of data-driven decision-making. For example, if an AI contractor for the DoD fails to maintain the integrity of their program and allows the model to use doctored or monitored data, then the contractor would be liable under the FCA for false certifications of cybersecurity and risk management compliance. Similarly, an HHS contractor would be liable if it misrepresents the accuracy of the model or conceals avenues for error or bias that materially affect CMS payment decisions, such as AI recommending or justifying inappropriate diagnostic codes.
While both examples mirror prior FCA cases regarding defense and healthcare fraud, they also display a growing tension in FCA litigation between technological complexity and legal accountability. AI models produce outputs through data-analytics from inputs government workers provide. Since no tangible goods are exchanged, the distinction between honest errors and actionable fraud begins to blur. In AI contracts, harm may manifest in subtle or delayed ways. Models could potentially return biased predictions or provide unreliable analytics that misinform decision making. The downstream consequences of a model’s flaws may be harder to identify. Since human decision-makers use AI outputs to guide their actions, rather than dictate them, defendants could argue that their judgement caused the harm rather than the AI model’s flaws.
Courts will soon have to define falsity in AI contexts. In prior FCA cases, falsity involved factors like misrepresented deliverables, inflated billing, or inadequate compliance. AI complicates defining the falsity of a claim in FCA cases. Relators will also face new challenges satisfying the scienter requirement as a contractor’s knowledge, deliberate ignorance, or reckless disregard for the falsity of their claim becomes harder to determine due to the autonomous nature of AI.
The autonomy of AI systems will make determining the intent of a defendant in FCA cases more complex. AI models’ opacity further complicates the issue. Many AI models are “black box” systems, meaning users, and often creators, cannot fully oversee the internal functions of the AI’s data-analysis nor reasoning for a given output. Where traditionally FCA cases analyzed intent through a given company’s internal communications or its employee’s actions, the layered corporate structures and technical teams responsible for the development and maintenance of a model may not fully know how exactly a deployed model evolves or produces results. Contractors could then reasonably argue that they were not aware of a model’s bias or its false outputs as they were emergent or the product of algorithmic drift rather than human influence.
Discovery in FCA cases involving AI will be exceptionally complex. In order to capture relevant information, AI contractors will need to supply the model architecture, training data, records of inputs and outputs, as well as all other relevant materials. Since AI models retrain and adjust to new data, when litigation arises, the model could feasibly not exist in the form it did during the relevant period of the case. As a result, preservation becomes essential for the relator’s ability to prove what was false during the time of contracting and payment. Disputes invoking trade secret and privacy protections for particular data sets will only serve to further complicate the process. These disputes will affect the scienter analyses as relators may have to rely on internal communications and requirements to determine if a defendant “knew” about flaws in a model’s performance.
Federal agencies accept a degree of uncertainty in AI performance while investing in the emergent technology. This uncertainty complicates the materiality element of FCA cases since a claim is material only if the government had refused payment had it known of the misrepresentation. When using the precedent set by United Health Services v. Escobar, courts will struggle to determine whether flaws in an AI model can meet the threshold for material misrepresentation of a good or service since the government may knowingly accept such a risk. A contractor’s false claims regarding an AI model’s function alone may not satisfy the FCA’s materiality requirement if the government implicitly consented to a measure of inaccuracy in the system’s outputs.
Federal agencies will need to strengthen contractual oversight and establish clear mechanisms for monitoring the use of AI. As the government develops the relevant policies over time, FCA litigation appears poised to be the proving ground for how the legal system will handle algorithmic accountability.
Henry Fina is a Project Analyst and Matthew P. Suzor is an Associate at Miller Shah LLP.
The views, opinions and positions expressed within all posts are those of the author(s) alone and do not represent those of the Program on Corporate Compliance and Enforcement (PCCE) or of the New York University School of Law. PCCE makes no representations as to the accuracy, completeness and validity or any statements made on this site and will not be liable any errors, omissions or representations. The copyright of this content belongs to the author(s) and any liability with regards to infringement of intellectual property rights remains with the author(s).