BRENNEN

Industries / Health / Healthcare / Medical billing

Compliance

Keeping bills honest and lawful: the laws against false claims and kickbacks, the audits that recheck paid claims, and the investigators who pursue fraud.

7

Milestones achieved, from 1986 to 2026.

2

To watch for: announced, not yet confirmed outside.

2

Open deliverables. Nobody has shipped them.

The road

AchievedWatch forOpen

Achieved

  1. 1986

    Triple damages for false claims

    The US False Claims Act of 1863, passed because suppliers were cheating the Union Army, made anyone who knowingly submitted a false claim to the government liable for double damages plus $2,000 a claim. Amendments in 1986 raised this to triple damages and $5,000 to $10,000 a claim, now adjusted for inflation. Under the Act's qui tam provisions a private person, the relator, can sue on the government's behalf and receives 15% to 25% of the recovery if the government joins the case, or 25% to 30% if it does not. A claim that results from a kickback may itself be false under the Act, by the Inspector General's account. The Justice Department obtained more than $2.9 billion in fraud and false claims settlements and judgments, across all sectors, in the year to 30 September 2024.

    justice.govoig.hhs.govjustice.gov

    Credit

    US Congress

  2. 1996

    Improper payments put at 14%

    An improper payment is one that should not have been made, or was made in the wrong amount, under the program's rules; it is not necessarily fraud. Auditing the financial statements of the Health Care Financing Administration, now CMS, the health department's Office of Inspector General estimated improper Medicare fee-for-service payments at $23.2 billion, 14% of the total, in fiscal year 1996, with a 95% confidence range of $17.8 billion to $28.6 billion, and $20.3 billion, 11%, in 1997. CMS's Comprehensive Error Rate Testing program now reviews a random sample of claims each year; it put the fiscal year 2026 rate at 7.32%, or $33.96 billion, and 16.67% for medical equipment and supplies.

    oig.hhs.govcms.gov

    Credit

    HHS Office of Inspector General under June Gibbs Brown; CMS

  3. 2003

    A $1.7 billion hospital settlement

    The US hospital company HCA, formerly Columbia/HCA, was accused of cost report fraud, overstating its costs in the annual cost reports hospitals file with Medicare, and of paying kickbacks to physicians. Its subsidiaries pleaded guilty in December 2000 and paid more than $840 million in criminal fines, civil restitution and penalties; on 26 June 2003 HCA agreed to pay $631 million more to settle civil claims, and $250 million to CMS. The Justice Department called the $1.7 billion total "by far the largest recovery ever reached by the government in a health care fraud investigation". Whistleblowers in the qui tam suits shared $151,591,500, which the department called the highest combined qui tam award it had paid.

    justice.gov

    Credit

    US Department of Justice; HHS Office of Inspector General; the relators

  4. 2005

    Auditors paid by findings

    Recovery audit contractors (RACs) recheck claims Medicare has already paid and are paid a share of what they recover, the first contingency fee contract in Medicare, by the Congressional Research Service's account. A pilot ran in California, Florida and New York from 2005 and in three more states from 2007. CMS reported that it corrected over $1 billion in improper payments, returned $693.6 million net to the trust funds and cost 20 cents per dollar collected. Hospital and physician groups objected that the fee rewarded aggressive denials, and a third of appealed findings were overturned. After the national program began, 437,524 backlogged RAC appeals were pending before administrative law judges on 30 September 2015, by the health department's own count in a court filing.

    LessonCMS called the validation contractor and the three-year limit on looking back, which it added for the national program, lessons of the pilot; the appeals backlog showed a contingency fee also moves cost to the providers and judges who handle disputed findings.

    cms.goveverycrsreport.comaha.org

    Credit

    US Congress (Medicare Modernization Act of 2003); CMS

  5. 2007

    Strike forces mine claims data

    Medicare Fraud Strike Force teams, first set up in March 2007, put the Office of Inspector General, the Justice Department, the FBI and local police together to find fraud in claims data and prosecute it quickly. By the Inspector General's count they had brought 3,315 criminal actions and 3,937 indictments by 31 March 2026. The 2026 national takedown, announced on 23 June 2026, charged 455 defendants in schemes the department says involved over $6.5 billion in false claims; one set of prosecutions began when the Justice Department's data analytics team spotted a spike in payments for allografts, tissue products taken from donors. The charges are allegations until proven.

    oig.hhs.govjustice.gov

    Credit

    HHS Office of Inspector General; US Department of Justice, Fraud Section

  6. 2023

    Australia reviews Medicare claiming

    After press reports of widespread misuse of Australia's Medicare, the government asked the health economist Pradeep Philip to review it in November 2022. His 2023 review put the likely cost of non-compliance at A$1.5 to 3 billion a year and found no evidence for the A$8 billion figure in some media reports, while warning that losses could grow to that size without better controls. It noted that five earlier reviews between 2016 and 2021 had flagged the losses, and made 23 recommendations, among them stronger governance, laws fit for purpose and better detection.

    Lessonin Philip's words, the number "should not be the main subject of debate"; the controls in the system should.

    hospitalhealth.com.au

    Credit

    Pradeep Philip; Department of Health and Aged Care

  7. 2026

    Added diagnoses cost $556 million

    Medicare Advantage plans, private plans that run Medicare benefits, are paid more for members with more recorded illnesses, a method called risk adjustment. On 14 January 2026 affiliates of Kaiser Permanente agreed to pay $556 million to resolve allegations that from 2009 to 2018 they pressed physicians to add diagnoses to patients' records through addenda, often months after the visit, to raise those payments. Two former employees who brought whistleblower suits share $95 million. The Justice Department states that the claims are allegations only and that there has been no determination of liability.

    justice.gov

    Credit

    US Department of Justice; HHS Office of Inspector General; relators Ronda Osinek and James M. Taylor

Watch for

  1. Every Medicare Advantage contract audited

    On 21 May 2025 CMS said it would audit all eligible Medicare Advantage contracts, about 550, each payment year instead of about 60, raise its medical coders from 40 to about 2,000 and use software to flag diagnoses that the records do not support, finishing the audits for 2018 to 2024 by early 2026. It cited federal estimates of about $17 billion a year in overbilling. On 25 September 2025 a federal court in Texas vacated the 2023 rule under which CMS planned to extrapolate audit findings, applying a sample's error rate to a whole contract; CMS appealed on 21 November 2025.

    Achieved whenCMS or the Inspector General reports audits completed for all eligible contracts in a payment year, with the amounts recovered, and the appeal is decided.

    cms.govcrowell.comcrowell.com

    Claimed by

    CMS, May 2025

  2. Fraud stopped before payment

    With the 2026 takedown the health department said its Inspector General had taken 25 actions seeking more than $10 billion in payments "that CMS caught and suspended before the funds were paid", and CMS's administrator said the agency is "deploying advanced data analytics" to freeze suspicious payments. The Justice Department's Fraud Section is to get computing space inside CMS's claims data repository to run analytics and AI tools.

    Achieved whenthe Government Accountability Office or the Inspector General audits the suspended payments and reports how many were confirmed improper and how many were later released to legitimate providers.

    justice.gov

    Claimed by

    US Department of Justice, HHS and CMS, June 2026

Open

  1. 01

    US claims with planted schemes

    A public, synthetic set of Medicare-style outpatient and medical equipment claims for invented patients and invented providers, matched in volume and mix to CMS's public provider-level summary files, with fraud schemes planted from patterns, not names, in the Justice Department's public case summaries: more than 24 hours of timed services in a day, services after death, code pairs that should have been billed as one, equipment never delivered, referral rings. Each planted claim carries the scheme's label and its dollar effect. Score detection methods on how much planted money they find when reviewing a fixed 1% of providers.

    The brief
    Credit · unclaimed
  2. 02

    Advisory opinions as a test

    A test of how well AI models apply the Anti-Kickback Statute, the federal criminal law against paying for referrals of patients covered by federal health programs. Take the facts from the Office of Inspector General's published advisory opinions, remove its conclusion, and ask each model whether the arrangement could draw sanctions and which safe harbor (a regulation that protects a practice meeting set conditions) applies; score the answers against the Inspector General's. Add an invented variant of each case with one fact changed, such as a fee tied to the number of referrals, answered separately by two health care lawyers.

    The brief
    Credit · unclaimed

Open 01 · the brief

US claims with planted schemes

Credit · unclaimed

The first to publish is credited on this line, and the deliverable moves from Open to Achieved.

What to build

A public, synthetic set of Medicare-style outpatient and medical equipment claims for invented patients and invented providers, matched in volume and mix to CMS's public provider-level summary files, with fraud schemes planted from patterns, not names, in the Justice Department's public case summaries: more than 24 hours of timed services in a day, services after death, code pairs that should have been billed as one, equipment never delivered, referral rings. Each planted claim carries the scheme's label and its dollar effect. Score detection methods on how much planted money they find when reviewing a fixed 1% of providers.

Why it is still open

US work uses real, restricted data. CMS's 2025 Crushing Fraud Chili Cook-Off gave its ten finalists Limited Data Sets under data use agreements and published no detection scores; academic studies label real Part B providers with the federal exclusion list, for example Herland, Bauder and Khoshgoftaar (Journal of Big Data, 2019). The public set most students use, posted on Kaggle by a user named Gupta, has inpatient and outpatient claims with a yes-or-no fraud flag for each provider, not each claim or scheme, and a 2025 survey by Rodolfo Rivas Matta of Florida Atlantic University notes that its source is unclear. CMS's synthetic claims files have no fraud labels and, in CMS's words, "very limited inferential research value". Synthetic benchmarks with planted fraud now exist for Japan (International Journal of Data Science and Analytics, October 2026) and for German inpatient claims (a 2024 simulator from Friedrich-Alexander-Universität Erlangen-Nürnberg), but not for US outpatient claims.

cms.govjournalofbigdata.springeropen.comlaccei.orgcms.govgithub.comlink.springer.com

Built with what exists

Synthea for invented patients; CMS's public provider-level files for volumes and prices; the NCCI and medically unlikely edit files to define code-pair and quantity violations; the 2026 takedown case summaries for scheme patterns; the Japan benchmark's design of labels tied to each change and reconciled in money; Python and open-source gradient-boosted tree and graph methods as baselines.

justice.gov

  1. First step

    Code 30 case summaries from the 2026 takedown into scheme patterns, then plant the "more than 24 hours in a day" pattern in a small Synthea-based claim set and test one simple detector.

  2. Done looks like

    at least 5 million claims from 20,000 invented providers, at least 10 scheme types each tied to a cited public case summary, labels at the level of each claim, baseline results for three methods, and the generator code, all published with a DOI.

Open 02 · the brief

Advisory opinions as a test

Credit · unclaimed

The first to publish is credited on this line, and the deliverable moves from Open to Achieved.

What to build

A test of how well AI models apply the Anti-Kickback Statute, the federal criminal law against paying for referrals of patients covered by federal health programs. Take the facts from the Office of Inspector General's published advisory opinions, remove its conclusion, and ask each model whether the arrangement could draw sanctions and which safe harbor (a regulation that protects a practice meeting set conditions) applies; score the answers against the Inspector General's. Add an invented variant of each case with one fact changed, such as a fee tied to the number of referrals, answered separately by two health care lawyers.

Why it is still open

The Inspector General has published redacted advisory opinions since its 1997 rule, as section 205 of HIPAA, the Health Insurance Portability and Accountability Act of 1996, requires, and each binds only the party that asked. They are public, reasoned rulings on specific arrangements, yet a search in November 2026 found no published test of AI models against them.

oig.hhs.govoig.hhs.gov

Built with what exists

the Inspector General's advisory opinions and its summary of the fraud and abuse laws; general AI models and open-weight models; a scoring rubric written with the two lawyers; Python.

oig.hhs.gov

  1. First step

    Take the 20 numbered opinions the Inspector General posted in 2026, strip their conclusions, and run two models on them.

  2. Done looks like

    at least 150 published opinions and 150 invented variants, the lawyers' agreement reported, at least four models scored on the conclusion and on naming the right safe harbor, with the prompts and answers published.