Across the healthcare industry, providers are increasingly relying on AI-assisted billing tools to automate medical coding, prior authorization workflows, and the submission of claims to Medicare, Medicaid, and other federal payors. Some vendors advertise “clean” claim rates exceeding 98%, meaning payors almost always accept submitted claims without further intervention or correction. But a high clean-claim rate is an operational metric, not a compliance safe harbor. It does not establish that the claim was properly coded, medically necessary, supported by documentation, or free from overpayment risk. The efficiency gains can be substantial, as can the heightened False Claims Act (“FCA”) exposure these systems can create. At the same time, many of these tools may improve consistency and reduce certain forms of human error when implemented and monitored appropriately. As AI continues to develop and becomes more widely integrated into healthcare billing, relators and prosecutors are likely to explore new avenues for evaluating, and litigating, how these tools are deployed, monitored, and overseen.

Like every billing system, AI-assisted billing tools can make mistakes. However, AI-assisted systems can scale those mistakes across thousands of claims, and the records those systems generate may make it easier for the government or relators to argue that a provider’s submission of those claims violated the FCA. The Department of Justice recovered more than $6.8 billion from FCA settlements and judgments in fiscal year 2025, the largest annual recovery in the statute’s history, with healthcare accounting for approximately $5.7 billion, or 83% of that total.[1] The same year saw 1,297 qui tam actions filed, the highest number ever recorded in a single year, with relator-initiated cases accounting for more than three-quarters of total FCA recoveries.[2] These numbers make clear that healthcare remains the government’s most significant FCA enforcement target, a trend that is unlikely to change anytime soon.

That enforcement environment is particularly relevant to AI-assisted billing tools. The DOJ-HHS False Claims Act Working Group has identified Medicare Advantage and manipulation of electronic health record systems to drive inappropriate utilization as priority FCA enforcement areas.[3] Organizations should therefore evaluate how AI-enabled documentation, coding, risk-adjustment, chart-review, and similar tools affect information used to support federal healthcare claims.

Audit logs, exception reports, override decisions, and similar records create a detailed evidentiary trail arguably showing what the provider knew, what issues were identified, and how those issues were handled. That is particularly significant because the FCA’s scienter standard extends beyond actual knowledge to include deliberate ignorance and reckless disregard. Importantly, the same system records that create evidentiary exposure are also powerful compliance infrastructure when used proactively, enabling providers to identify issues faster, extend staff capacity, and demonstrate a culture of oversight to regulators. The question is whether a provider is positioned to use those records defensively or will instead face them as evidence in an enforcement action.


The Knowledge Standard and What It Reaches

The FCA imposes liability on anyone who knowingly submits a false or fraudulent claim to the federal government, but the statute’s knowledge standard reaches beyond actual intent, encompassing deliberate ignorance and reckless disregard, both of which are substantially lower thresholds. The Supreme Court’s 2023 decision in United States ex rel. Schutte v. SuperValu Inc.[4] reinforces this point: the FCA’s scienter inquiry turns on what the defendant actually understood and believed at the time claims were submitted, not whether its conduct was objectively reasonable. For providers using AI-assisted billing tools, that means the relevant question is what the provider knew about the system’s operation, what risks were identified, and how it responded.[5]

The FCA does not impose strict liability for every inaccurate AI-assisted claim. A plaintiff must establish a false or fraudulent claim that was material to the government’s payment decision and must satisfy the statutory knowledge standard. That said, a provider who submits a materially false claim using an AI tool could face FCA exposure, even if the provider did not specifically intend to defraud the government. This is because automation relocates decision-making rather than eliminating it. A provider using an automated claims process may still be seen as deciding how that process operates, what oversight applies, which safeguards exist, and how exceptions are handled. In many cases, the tool may highlight those choices, generating a contemporaneous record of what the provider arguably understood at the time claims were submitted.

Where the Risk Actually Arises

The FCA framework has significant implications for providers deploying AI-assisted billing systems. The issue is often not whether the technology made a mistake in isolation, but whether the provider organization submitted materially false claims after ignoring warning signs, failing to implement meaningful oversight, or continuing to use a workflow despite known risks associated with the system’s operation. In practice, that exposure can arise in several common scenarios:

(1) A provider deploys an AI billing tool to maximize reimbursement, but without implementing sufficient guardrails, such as human oversight, model validation, or auditing protocols. Model outputs can trend toward higher-acuity codes or broader service classifications, producing systematic coding patterns at scale. Prompts embedded in clinical workflows can shape documentation in ways that drive reimbursement outcomes. What might once have been an isolated billing error may instead recur across thousands of claims when driven by automation. Under SuperValu, prosecutors and relators are likely to focus on the oversight, escalation, and visibility decisions a provider made when automating claim submissions—including whether contrary guidance, audit findings, exception reports, user complaints, or other warning signs were identified but not meaningfully addressed.

(2) A provider receives alerts of issues from the AI tool but fails to fix them. AI billing systems generate persistent, structured records. Audit trails, model outputs, override logs, exception reports, and confidence scores are commonly captured.[6] Those records allow reconstruction of how claims were generated and how the organization responded, or failed to respond, to identified issues. Internal users can adjust confidence thresholds, suppress flags, or bypass exception queues to increase throughput. Moreover, in many cases, the evidentiary record created by AI billing systems is more detailed than in traditional billing environments, and can equip relators, who are typically insiders with access to these records, or the government with powerful evidence of reckless disregard.

The Government Will Hold Providers Responsible Regardless of Vendor Agreements

Regardless of what any vendor agreement says, the government will treat providers as responsible for claim accuracy. Providers should build their oversight posture around that reality rather than around what their vendor agreement does or does not promise. In fact, many vendor contracts explicitly place full compliance responsibility on providers, often stating the customer is solely responsible for regulatory compliance or disclaiming any warranty that services will be error-free. In an enforcement context, those contractual allocations are not merely boilerplate. They serve to document how responsibility was understood and accepted at the time the tool was deployed.

Some agreements go even further, requiring the provider to indemnify the vendor against claims arising from billing errors or regulatory violations tied to use of the platform. The indemnification clauses in standard vendor agreements are almost always one-way: the vendor is protected from the provider’s mistakes, but the provider receives no reciprocal protection. In a False Claims Act investigation, the government would point to this language as a written acknowledgment that the provider, not the tool, was responsible for claim accuracy, and that the provider was on notice of the platform’s limitations before the first claim went out. Under a reckless disregard standard, that framing has teeth.

Building AI-Assisted Billing into a Durable Compliance Framework

Providers using AI-assisted billing tools should evaluate these systems with the understanding that regulators and relators will scrutinize not only claim outcomes, but also the provider’s oversight decisions, escalation procedures, and internal records surrounding the tool’s use. Providers that build robust practices now are not only reducing enforcement risk; they are affirmatively positioning themselves to use AI as a durable compliance asset, extending staff capacity, identifying issues faster, and demonstrating to regulators a culture of proactive oversight. At a minimum, providers should consider whether they can clearly demonstrate the following:

  • A designated compliance owner for AI billing oversight who has defined escalation authority and is responsible for ensuring that the organization understands how the platform functions operationally, including how outputs are generated, escalated, reviewed, and incorporated into claims submission workflows.
  • Periodic substantive review of override and exception decisions (not just logging them), along with meaningful oversight mechanisms, including validation procedures, auditing protocols, and monitoring of override activity, exception handling, and other indicators of potential billing risk.
  • A baseline post-implementation coding audit to compare AI-driven patterns against prior practice, along with evaluation of how AI-generated records, audit logs, and system outputs are preserved and whether existing retention practices adequately capture the evolving nature of AI-generated data.
  • A defined process for evaluating vendor model updates or retraining before they go live on claims, along with review of vendor agreements, disclaimers, and risk allocations with the understanding that they may later be examined by regulators or relators as evidence of how responsibility for billing accuracy was allocated and understood internally.

In many cases, the question will not be whether an AI-assisted billing system made a mistake. The question will be whether the organization submitted materially false claims, what the provider knew about the risk, what the system revealed internally, and how the provider responded once those issues surfaced. Providers that evaluate their AI billing practices through that lens before a subpoena or civil investigative demand arrives are not merely managing downside risk. They are building the infrastructure to use AI as a genuine compliance advantage, one that regulators will recognize and that will serve the organization well into a future where these tools are only going to become more central to healthcare operations.

For questions about this topic and its implications for companies developing or deploying AI, contact the authors or members of the firm’s Artificial Intelligence team.

[1] U.S. Dep’t of Justice, Office of Pub. Affairs, False Claims Act Settlements and Judgments Exceed $6.8B in Fiscal Year 2025 (press release), https://www.justice.gov/opa/pr/false-claims-act-settlements-and-judgments-exceed-68b-fiscal-year-2025 (Jan. 16, 2026).

[2] Id.

[3] U.S. Dep’t of Justice, Office of Pub. Affairs, DOJ-HHS False Claims Act Working Group (press release), https://www.justice.gov/opa/pr/doj-hhs-false-claims-act-working-group (July 2, 2025).

[4] 598 U.S. 739 (2023).

[5] Id. at 757. 

[6] See, e.g., QuickIntell, Healthcare Compliance Audit Survival Guide: How AI Creates an Airtight Audit Trail, https://quickintell.com/compliance/audit-survival-guide (last visited Apr. 1, 2026); see also Simeon Petrov & Jiban Khuntia, AI in Revenue Cycle Management (RCM) and Medical Claims Processing, 10 Health Admin. Rsch. Consortium 5–6 (Fall 2025), https://business.ucdenver.edu/sites/default/files/attached-files/harb-10-2-ai_in_revnue_cycle_managment_rcm.pdf.