GAO-25-108412: Fraud and Improper Payments: Data Quality and a Skilled Workforce Are Essential for Unlocking the Benefits of Artificial Intelligence

Summary

GAO-25-108412, Fraud and Improper Payments: Data Quality and a Skilled Workforce Are Essential for Unlocking the Benefits of Artificial Intelligence, the statement of Sterling Thomas, Chief Scientist, for testimony before the Joint Economic Committee, U.S. Congress, on April 9, 2025. It describes actions Congress and agencies can take against fraud and improper payments without AI, such as making permanent the Social Security Administration's authority to share its full death data with Treasury's Do Not Pay system. It states that AI tools require high-quality data and refers to GAO's AI Accountability Framework for Federal Agencies and Other Entities. It reports a severe shortage of federal staff with AI expertise.

Summary drafted by a model from the document's text below and checked by script against that text before publication. It is a navigation aid, not a reading of what the document proves. Where AI is used

Cite this document

“GAO-25-108412: Fraud and Improper Payments: Data Quality and a Skilled Workforce Are Essential for Unlocking the Benefits of Artificial Intelligence.” Original at www.gao.gov. Archived at Pandemic Darlings, https://pandemicdarlings.com/source-documents/gao-25-108412-fraud-and-improper-payments-data-quality-and-a-skilled-w-7936586eec/, SHA-256 76dd5b6ec59d08a4f861f44367fd632342200b2ea047e7076bfaa2a57765b38b.

Full text

United States Government Accountability Office
Testimony
Before the Joint Economic Committee,
U.S. Congress

FRAUD AND IMPROPER
For Release on Delivery
Expected at 2:30 p.m. ET
Wednesday, April 9, 2025

PAYMENTS
Data Quality and a Skilled
Workforce Are Essential
for Unlocking the Benefits
of Artificial Intelligence
Statement of Dr. Sterling Thomas, Chief Scientist

GAO-25-108412
April 2025

FRAUD AND IMPROPER PAYMENTS

Highlights Data Quality and a Skilled Workforce Are Essential for
Unlocking the Benefits of Artificial Intelligence
Highlights of GAO-25-108412, a testimony
before the Joint Economic Committee,
U.S. Congress

Why GAO Did This Study What GAO Found
GAO has reported that fraud and The federal government has many existing tools and resources to help agencies
improper payments are estimated to combat fraud and improper payments. GAO has recommended improvements in
have cost taxpayers trillions of dollars. these areas. For example, Congress could make permanent the Social Security
These issues also impact the integrity Administration’s authority to share its full death data with the Department of the
of federal programs and erode public Treasury’s Do Not Pay system.
trust. Improper payments are
payments that should not have been Programs Reporting the Largest Percentage of Estimated Improper Payments in Fiscal Year
made or that were made in the wrong 2024
amount. Fraud involves obtaining
something of value through willful
misrepresentation.
The advancement of AI presents both
opportunities and challenges for
combatting fraud and improper
payments in the federal government.
This testimony describes 1) actions
Congress and agencies can take to
combat fraud and improper payments
without the use of AI, 2) opportunities
and challenges for using AI to combat
fraud and improper payments, and 3)
workforce challenges in the use of AI in
the federal government.

Note: See full report for details of payment estimates.

Artificial intelligence (AI) and other innovative technologies have the potential to
enhance efforts to combat fraud and improper payments. However, these tools
require high-quality data. Introducing insufficient, unrelated, or bad data will make
an AI model less reliable. A system that produces errors will also erode trust in
the use of AI to detect fraud. GAO’s AI Accountability Framework for Federal
Agencies and Other Entities includes key practices for ensuring data are high-
quality, reliable, and appropriate for the intended purpose. Another potential step,
which GAO recommended in 2022, is to establish a permanent analytics center
of excellence focused on fraud and improper payments. Should such a center be
realized, it is likely that an AI-based tool would be a key component.
An AI-ready workforce is another essential requirement if the federal government
is to use this tool in the fight against fraud and improper payments. For decades,
however, GAO has identified mission-critical gaps in federal workforce skills and
expertise in science, technology, engineering, and mathematics. More
specifically, there is a severe shortage of federal staff with AI expertise. GAO has
reported that improvements may be hampered by uncompetitive compensation
and the lengthy federal hiring process.
For more information, contact Sterling Thomas
at ThomasS2@gao.gov.

United States Government Accountability Office
Letter Letter

Chairman Schweikert, Ranking Member Hassan, and Members of the
Committee:

Thank you for the opportunity to discuss using artificial intelligence (AI) to
combat fraud and improper payments.

GAO has reported that improper payments and fraud are estimated to
have collectively cost taxpayers trillions of dollars. In addition, these
issues impact the integrity of federal programs, and erode public trust. 1
Improper payments are payments that should not have been made or that
were made in the wrong amount, typically overpayments. 2 Fraud involves
obtaining something of value through willful misrepresentation. 3 We
estimated that the federal government loses between $233 billion and
$521 billion annually from fraud, based on fiscal year 2018 to 2022 data. 4
Since fiscal year 2003, cumulative improper payment estimates by
executive branch agencies have totaled about $2.8 trillion, but are almost
certainly greater, in part because agencies have not reported estimates
for some programs as required. For example, last year we reported that
agencies failed to report fiscal year 2023 improper payment estimates for

1GAO, Improper Payments: Agency Reporting of Payment Integrity Information,
GAO-25-107552, (Washington, D.C.: Jan. 23, 2025). GAO, Improper Payments and
Fraud: How They Are Related but Different, GAO-24-106608, (Washington, D.C.: Dec. 7,
2023).
2An improper payment is defined by law as any payment that should not have been made
or that was made in an incorrect amount (including overpayments and underpayments)
under statutory, contractual, administrative, or other legally applicable requirements. It
includes any payment to an ineligible recipient, any payment for an ineligible good or
service, any duplicate payment, any payment for a good or service not received (except
for such payments where authorized by law), and any payment that does not account for
credit for applicable discounts. 31 U.S.C. § 3351(4). Executive agency heads are required
periodically to review the programs and activities they administer and make an estimation
of each program or activity’s improper payments. If in conducting such a review an agency
head is unable to discern whether a payment was proper because of insufficient or lack of
documentation, that payment must also be included in the improper payment estimate. 31
U.S.C. § 3352(a),(c).
3Fraud can sometimes involve benefits that do not result in direct financial loss to the
government (such as passport fraud). Improper payments, fraud, and fraud risk are
related but distinct concepts. While unintentional error may cause improper payments,
fraud involves obtaining something of value through willful misrepresentation. Whether an
act is fraudulent is determined through the judicial or other adjudicative system. Fraud risk
exists when individuals have an opportunity to engage in fraudulent activity.
4GAO. Fraud Risk Management: 2018-2022 Data Show Federal Government Loses an
Estimated $233 Billion to $521 Billion Annually to Fraud, Based on Various Risk
Environments, GAO-24-105833, (Washington, D.C.: Apr 16, 2024).

Page 2 GAO-25-108412
nine risk-susceptible programs. 5 Figure 1 shows that about 75 percent
($121 billion) of total government-wide estimated improper payments
reported for fiscal year 2024 are concentrated in five program areas.

Figure 1: Programs Reporting the Largest Percentage of Improper Payments Estimates in Fiscal Year 2024

Note: Improper payment estimates displayed in the figure include both improper and unknown
payments. Executive agency estimates of improper payments treat as improper any payments whose
propriety cannot be determined due to lacking or insufficient documentation. 31 U.S.C §.
3352(c)(2)(A).

The advancement of AI presents both opportunities and challenges for
combatting improper payments and fraud in the federal government.
Today I will discuss 1) actions Congress and agencies can take to
combat fraud and improper payments without the use of AI, 2)
opportunities and challenges for using AI to combat fraud and improper
payments, and 3) workforce challenges in the use of AI in the federal
government. To complete this work, we reviewed GAO’s body of work on
improper payments, fraud, and artificial intelligence. Specifically, we
assessed reports on federal financial management, fraud prevention,

5GAO, Improper Payments: Key Concepts and Information on Programs with High Rates
or Lacking Estimates, GAO-24-107482 (Washington, D.C.: June 27, 2024).

Page 3 GAO-25-108412
GAO’s High Risk List, and artificial intelligence. Appendix I shows
selected GAO work on AI.

We performed the work on which this testimony is based in accordance
with all applicable sections of GAO’s Quality Assurance Framework.

While eliminating all fraud is not a realistic goal, the federal government
Congress and has many tools and resources already in place to help agencies prevent
Agencies Could fraud prior to funds distribution and to combat fraud and improper
payments once they have occurred. 6 Agency and congressional action on
Improve Existing our recommendations for improvements in these areas could enhance
Systems to Combat these efforts.
Fraud and Improper One key resource for preventing improper payments and fraud is the
Payments Department of the Treasury’s Do Not Pay system, which consolidates
data on ineligible entities. The Social Security Administration (SSA)
currently shares its full death data with Do Not Pay, but its authority to do
so expires next year. We have recommended that Congress consider
making it permanent. 7 In January 2025, Treasury reported recovering $31
million in payments to deceased individuals during a 5-month period on
the pilot with the SSA Full Death Master file. 8

For fraud risk management, preventive activities generally offer the most
cost-efficient use of resources, since they enable managers to avoid a
costly and inefficient “pay-and-chase” model. Managers of federal
programs maintain the primary responsibility for managing fraud risk.
Since 2016, agencies have been required to adhere to leading practices
in GAO’s Fraud Risk Framework to effectively design, implement, and

6Agency Inspectors General play an important role in investigating instances of fraud in
their respective agencies. Per the Inspector General Act of 1978, as amended, the
mission of agency Inspectors General includes preventing and detecting fraud, waste, and
abuse. 5 U.S.C. § 402(b). However, often by the time an Office of Inspector General
detects fraud or an improper payment that has already occurred, it is challenging to get
the money back.
7In March 2022, we recommended that Congress amend the Social Security Act to
accelerate and make permanent the requirement for SSA to share its full death data with
the Treasury’s Do Not Pay system. GAO, Emergency Relief Funds: Significant
Improvements Are Needed to Ensure Transparency and Accountability for COVID-19 and
Beyond, GAO-22-105715 (Washington, D.C.: Mar. 17, 2022).
8“Treasury Data Pilot Prevents and Recovers $31 Million in Payments to Deceased

Individuals During a Five-Month Period,” U.S. Department of the Treasury, January 15,
2025, https://home.treasury.gov/news/press-releases/jy2784.

Page 4 GAO-25-108412
operate an internal control system that addresses fraud risks. 9 A leading
practice in the Fraud Risk Framework involves designing and
implementing data analytic controls to prevent and detect fraud. 10 Our
prior work has highlighted areas in which federal agencies need to take
additional actions to help ensure they are effectively managing fraud risks
consistent with leading practices, such as using data analytics to better
manage fraud risk. Specifically, from July 2015 through August 2023, we
made 47 recommendations to federal agencies in this area. These
include recommendations to design and implement data-analytics
activities to prevent and detect fraud, such as using data matching to
verify self-reported information. Of the 47 recommendations, a little more
than half have been implemented as of April 2024. To further help
agencies build prevention-focused anti-fraud efforts, GAO developed its
web-based Antifraud Resource, which provides interactive tools and
resources for understanding and combatting fraud. 11

For example, our Antifraud Resource provides curated resources that can
help agencies identify case examples, guidance, and other tools for
combatting various types of fraud in federal programs (see fig. 2).
Implementing our recommendations and using our resources can enable
agencies to carry out their missions and better protect taxpayer dollars
from fraud.

9GAO, A Framework for Managing Fraud Risks in Federal Programs, GAO-15-593SP
(Washington, D.C.: July 28, 2015)
10OMB’s Circular No. A-123, Management’s Responsibility for Enterprise Risk
Management and Internal Control, directs executive agencies to adhere to the Fraud Risk
Framework’s leading practices as part of their efforts to effectively design, implement, and
operate an internal control system that addresses fraud risks.
11GAO Fraud Ontology Version 1.1 (Washington, D.C.: Nov. 15, 2024),
https://gaoinnovations.gov/antifraud_resource/howfraudworks.

Page 5 GAO-25-108412
Figure 2: Five Categories of Antifraud Resources

New innovations like AI-enabled tools can perform some of the tasks
Opportunities and required to combat fraud and improper payments in the federal
Challenges for Using government. However, these tools require sufficient high-quality data, an
understanding of risks, and staff with knowledge of how to use AI.
AI to Combat Introducing insufficient, unrelated, or bad data will make an AI model less
Improper Payments reliable, and maintaining a “human in the loop” is vital to ensuring
oversight of the data and processes. A system that produces errors will
and Fraud also erode trust in AI to detect fraud.

AI, in general, refers to computer systems that can solve problems and
perform tasks that have traditionally required human intelligence. Machine
learning is a subset of AI that underpins many of the recent improvements
in the field and could be used to detect fraud and improper payments.
Many federal agencies are using or planning to use machine learning,
including to support detecting improper payments.

In recognition of the opportunities and risks of AI, in 2021, we published
Artificial Intelligence: An Accountability Framework for Federal Agencies

Page 6 GAO-25-108412
and Other Entities (see fig. 3). 12 We identified 31 key practices to help
ensure accountability and responsible AI use by federal agencies and
other entities, which could include use for combatting fraud and improper
payments. These key practices include ensuring systems are
documented, designed, and governed appropriately for their intended
uses; ensuring data quality; and recruiting and retaining personnel with
the necessary multidisciplinary skills and experiences.

Figure 3: GAO Artificial Intelligence (AI) Accountability Framework

12GAO, Artificial Intelligence: An Accountability Framework for Federal Agencies and
Other Entities, GAO-21-519SP, (Washington, D.C.: June 30, 2021).

Page 7 GAO-25-108412
Machine learning algorithms work by identifying statistical relationships
between inputs and outputs from a training dataset. Training data can
include numbers, images, or text. Training is the process of feeding
training data through the algorithm until it identifies statistical relationships
of interest.

Such a model might improve detection or prevention of fraud and
improper payments by revealing anomalous patterns, behaviors, and
relationships with a speed and scale that was not possible before. 13 But
machine learning models can also pose risks. For example, they may fail
to detect improper payments (false negatives) and they can erroneously
identify legitimate payments as improper (false positives). False positives,
in turn, can delay or deny payments to rightful recipients, such as small
businesses and beneficiaries of Social Security and Medicare.

Critical to mitigating such risk is training the model on high-quality data. A
common phrase among AI developers is “garbage in, garbage out,”
meaning that poor data will give poor results. This axiom also applies to
AI for detecting fraud and improper payments. In our AI Accountability
Framework, we identified five key practices to help entities use
appropriate data for developing AI models. For example, entities should
document sources and origins of training data and ensure that the data
are incorporated into the most appropriate AI model.

For detecting fraud and improper payments, machine learning systems
could require training data in the form of payments labeled as one of
three categories: accurate, improper without fraud, or fraud. Labeling
historical data incorrectly could lead to false results. If these become too
numerous, agencies will spend more time identifying AI’s mistakes than
they will save compared with traditional detection methods. The training
data could also be adulterated by “data poisoning,” a process by which
someone changes the data, which changes the behavior of a system.

Technology can be used to identify targets of opportunity that allow the
government agencies and Inspectors General to probe further into
suspected fraud. AI tools can be a useful technology and will further
evolve. But we need solid, reliable “ground truth” data and a human in the
loop to ensure data reliability and appropriate application of the
technology. AI does not replace the professional judgment of experienced

13 Taka Ariga, “Artificial Intelligence Creates New Opportunities to Combat Fraud.”
International Journal of Government Auditing, Summer 2020 Edition.

Page 8 GAO-25-108412
staff in detecting potentially fraudulent activities. While AI can sift through
large volumes of data, human intelligence is still an essential element for
choosing appropriate actions and technology tools.

Whether government data will reliably provide this ground truth is unclear.
Such data vary in quality and standards. We have recommended
numerous improvements to federal agencies, such as data mining and
other analytic practices to ensure data are sufficient for fraud risk
analysis. In our 2023 survey on fraud risk management, federal agencies
told us that access to data to look for fraud indicators was a challenge. 14

To improve the use of data analytics in identifying fraud and improper
payments, we recommended in 2022 that Congress establish a
permanent analytics center of excellence, similar to the Pandemic
Analytics Center of Excellence (PACE). 15 Should such a center be
realized, it is likely that an AI-based tool would be a key component.

An AI-ready workforce is another essential requirement if the federal
An AI-Ready Federal government is to use this tool in the fight against fraud and improper
Workforce Is payments. For decades, however, we have identified mission-critical gaps
in federal workforce skills and expertise in science, technology,
Essential for engineering, and mathematics. More specifically, there is a severe
Innovation shortage of federal staff with AI expertise.

One option Congress has considered is establishing a new digital
services academy—like the military academies—to train future workers. 16
In 2021, we convened technology leaders from government, academia,
and nonprofits to discuss such an academy and related issues. Among
their comments:

• Current federal digital staff compensation is not competitive.

14 GAO, Fraud Risk Management: Agencies Should Continue Efforts to Implement
Leading Practices, GAO-24-106565 (Washington, D.C.: Nov. 01, 2023).
15GAO-22-105715. And, according to a March 31, 2025, press release from PACE,
investigators used the PACE data analytics platform to untangle the electronic trail that
fraudsters left behind in the amount of $109 million. “Statement from PRAC Chair Michael
E. Horowitz Following Guilty Pleas in $109 Million Pandemic Fraud Investigation
Supported by the Pandemic Analytics Center of Excellence” PandemicOversight.gov,
March 31, 2025, https://pandemicoversight.gov/news/articles/statement-from-prac-chair-
following-guilty-pleas-in-109-million-pandemic-fraud-investigation-supported-by-pace.
16GAO, Digital Services: Considerations for a Federal Academy to Develop a Pipeline of
Digital Staff, GAO-22-105388, (Washington, D.C.: Nov. 19, 2021).

Page 9 GAO-25-108412
• Many digital staff may not be willing to endure the lengthy federal
hiring process.
• An academy might best focus on master’s degrees because agencies
need staff with advanced skills.

Chairman Schweikert, Ranking Member Hassan, and Members of the
Committee, this completes my prepared statement. I would be pleased to
respond to any questions that you may have at this time.

Sterling Thomas, Chief Scientist, ThomasS2@gao.gov.
GAO Contact and
Staff GAO staff who made key contributions to this testimony are Lisa Gardner
(Assistant Director), Claire McLellan (Analyst-in-Charge), Jenique
Acknowledgments Meekins, Joseph Rando, and Ben Shouse. With contributions from Seto
Bagdoyan, Virginia Chanley, Alex Gromadzki, Michael Hoffman, Hannah
Padilla, Rebecca Shea, Jared Smith, Andrew Stavisky, Kevin Walsh, and
Candice Wright.

Page 10 GAO-25-108412
Appendix I: A Selection of Artificial
Appendix I: A Selection of Artificial Intelligence
Work from GAO

Intelligence Work from GAO

Page 11 GAO-25-108412
Appendix I: A Selection of Artificial Intelligence
Work from GAO

Page 12 GAO-25-108412
This is a work of the U.S. government and is not subject to copyright protection in the
United States. The published product may be reproduced and distributed in its entirety
without further permission from GAO. However, because this work may contain
copyrighted images or other material, permission from the copyright holder may be
necessary if you wish to reproduce this material separately.
The Government Accountability Office, the audit, evaluation, and investigative
GAO’s Mission arm of Congress, exists to support Congress in meeting its constitutional
responsibilities and to help improve the performance and accountability of the
federal government for the American people. GAO examines the use of public
funds; evaluates federal programs and policies; and provides analyses,
recommendations, and other assistance to help Congress make informed
oversight, policy, and funding decisions. GAO’s commitment to good government
is reflected in its core values of accountability, integrity, and reliability.

The fastest and easiest way to obtain copies of GAO documents at no cost is
Obtaining Copies of through our website. Each weekday afternoon, GAO posts on its website newly
GAO Reports and released reports, testimony, and correspondence. You can also subscribe to
GAO’s email updates to receive notification of newly posted products.
Testimony
Order by Phone The price of each GAO publication reflects GAO’s actual cost of production and
distribution and depends on the number of pages in the publication and whether
the publication is printed in color or black and white. Pricing and ordering
information is posted on GAO’s website, https://www.gao.gov/ordering.htm.
Place orders by calling (202) 512-6000, toll free (866) 801-7077, or
TDD (202) 512-2537.
Orders may be paid for using American Express, Discover Card, MasterCard,
Visa, check, or money order. Call for additional information.

Connect with GAO on X, LinkedIn, Instagram, and YouTube.
Connect with GAO Subscribe to our Email Updates. Listen to our Podcasts.
Visit GAO on the web at https://www.gao.gov.

Contact FraudNet:
To Report Fraud,
Website: https://www.gao.gov/about/what-gao-does/fraudnet
Waste, and Abuse in
Automated answering system: (800) 424-5454
Federal Programs
Sarah Kaczmarek, Managing Director, Media@gao.gov
Media Relations

A. Nicole Clowers, Managing Director, CongRel@gao.gov
Congressional
Relations
https://www.gao.gov/about/contact-us
General Inquiries

Please Print on Recycled Paper.

File and source

File: gao-25-108412.pdf
Size: 2,324,445 bytes
SHA-256: 76dd5b6ec59d08a4f861f44367fd632342200b2ea047e7076bfaa2a57765b38b
Our copy: gao-25-108412.pdf
Original: www.gao.gov