GAO-26-108850: Fraud and Improper Payments: Data Quality and a Skilled Workforce Are Essential for Realizing Artificial Intelligence’s Benefits

Summary

GAO-26-108850, Fraud and Improper Payments: Data Quality and a Skilled Workforce Are Essential for Realizing Artificial Intelligence's Benefits, the statement of Sterling Thomas, Chief Scientist, for testimony before the Subcommittee on Government Operations, Committee on Oversight and Government Reform, House of Representatives, on January 13, 2026. It states that GAO estimated the federal government loses between $233 billion and $521 billion annually due to fraud, based on data for fiscal years 2018 through 2022. It states that, by implementing a GAO recommendation, the Small Business Administration identified $4.7 billion in Paycheck Protection Program loans made before September 2020 that went to ineligible recipients or were used for unauthorized purposes. It addresses data quality and workforce needs for using AI against fraud.

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-26-108850: Fraud and Improper Payments: Data Quality and a Skilled Workforce Are Essential for Realizing Artificial Intelligence’s Benefits.” Original at www.gao.gov. Archived at Pandemic Darlings, https://pandemicdarlings.com/source-documents/gao-26-108850-fraud-and-improper-payments-data-quality-and-a-skilled-w-88c7053d37/, SHA-256 32bd8516ec04714e20f9ee3df7967517156c4453464aa2ab9ca76226bf797e7e.

Full text

United States Government Accountability Office
Testimony
Before the Subcommittee on
Government Operations, Committee
on Oversight and Government Reform,
House of Representatives
FRAUD AND IMPROPER
For Release on Delivery
Expected at 2:00 p.m. ET
Tuesday, January 13, 2026

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

GAO-26-108850
FRAUD AND IMPROPER PAYMENTS

Highlights Data Quality and a Skilled Workforce Are Essential for
Realizing Artificial Intelligence’s Benefits
GAO-26-108850 January 2026
A testimony before the Subcommittee on Government Operations, Committee on Oversight and Government Reform, House of
Representatives. For more information, contact: Sterling Thomas at ThomasS2@gao.gov

What GAO Found Why GAO Did This Study
The federal government has tools and resources to help agencies combat fraud The distinct, yet interrelated, problems
and improper payments. GAO has recommended improvements to the use of of fraud and improper payments are
these tools and resources. For example, Congress should consider making long-standing and pervasive. Fraud
permanent the Social Security Administration’s requirement to share its full death involves obtaining something of value
data with the Do Not Pay system to help prevent fraud and improper payments. through willful misrepresentation.
Further, GAO has identified leading practices for managing fraud risks at federal Improper payments are payments that
agencies and has made recommendations to agencies to implement these should not have been made or that
practices. For example, in 2024, GAO recommended that the Department of were made in the wrong amount. GAO
estimated that the federal government
Defense revise its Fraud Risk Management Strategy to include data analytics as
loses between $233 billion and $521
a method to address fraud. Further, by implementing GAO’s recommendation,
billion annually due to fraud, based on
the Small Business Administration identified $4.7 billion in loans from the data for fiscal years 2018 through 2022.
Paycheck Protection Program that were made before September 2020 and that Since fiscal year 2003, cumulative
went to ineligible recipients or were used for unauthorized purposes. improper payment estimates reported
by executive branch agencies have
Programs Reporting the Largest Estimates of Improper Payments in Fiscal Year
2024
totaled about $2.8 trillion. In fiscal year
2024, five programs accounted for most
improper payments. These issues
impact the integrity of federal programs
and erode public trust.
The advancement of AI and other
innovative technologies presents
opportunities and challenges for
combatting fraud and improper
payments. This statement discusses (1)
examples of tools and resources
Congress and federal agencies can
leverage to improve existing efforts to
combat fraud and improper payments
without the use of AI, (2) opportunities
and challenges for using data analytics
and AI to combat fraud and improper
payments, and (3) challenges in
Artificial intelligence (AI) and data analytics have the potential to enhance efforts developing an AI-ready federal
to combat fraud and improper payments but also have challenges. For example, workforce. This statement is based on a
body of work GAO issued between
data analytics and AI could help agencies sift through large volumes of data.
2015 and 2025 including GAO-25-
However, agencies need solid, reliable data and a human in the loop to ensure
108172, GAO-25-107508,
data reliability and appropriate application of the technology. GAO’s AI GAO-21-519SP, and GAO-25-107653.
Accountability Framework for Federal Agencies and Other Entities includes key
practices for ensuring data used in AI systems are high quality, reliable, and What GAO Recommends
appropriate for the intended purpose. Further, to improve the use of data
analytics in identifying fraud and improper payments, GAO recommended in GAO has made numerous
2022 that Congress establish a permanent analytics center of excellence. recommendations to federal agencies
and matters for Congressional
Additionally, the federal government requires an AI-ready workforce if AI is to consideration to address fraud and
help combat fraud and improper payments. However, GAO has identified improper payments in federal programs.
mission-critical gaps in science, technology, engineering, and mathematics skills
within the federal workforce and has reported on challenges agencies face in
attracting and developing individuals with AI expertise.

United States Government Accountability Office
Chairman Sessions, Ranking Member Mfume, and Members of the
Subcommittee:

Thank you for the opportunity to discuss artificial intelligence (AI) and
data analytics, as well as other resources to combat fraud and improper
payments in federal programs.

We have a long history of auditing fraud and improper payments, which
we and executive branch agencies estimated to have collectively cost
taxpayers trillions of dollars. In addition, these issues impact the integrity
of federal programs and erode public trust. 1 We have issued numerous
reports, guides, and recommendations about preventing fraud and
improper payments. While innovative tools like AI could help prevent and
detect fraud, as we reported in April 2025, these tools require careful
implementation, high-quality data, and a workforce trained to use them. 2

Fraud involves obtaining something of value through willful
misrepresentation. 3 Improper payments are payments that should not
have been made or that were made in the wrong amount, typically
overpayments. 4 In 2024, we estimated that the federal government loses
between $233 billion and $521 billion annually from fraud, using data 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).
2GAO, Fraud and Improper Payments: Data Quality and a Skilled Workforce Are Essential
for Unlocking the Benefits of Artificial Intelligence, GAO-25-108412 (Washington, D.C.:
Apr. 9, 2025).
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.
4An 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).

Page 2 GAO-26-108850
fiscal years 2018 through 2022. 5 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. 6
For example, in 2025 we reported that agencies did not report required
estimates for several risk-susceptible programs, including the Department
of Health and Human Services’ Temporary Assistance for Needy
Families, the Department of Housing and Urban Development’s Office of
Public and Indian Housing’s Tenant-Based Rental Assistance, and the
Small Business Administration’s Shuttered Venue Operators Grant
programs. 7 Figure 1 shows that about 75 percent ($121 billion) of total
government-wide estimated improper payments reported for fiscal year
2024 were concentrated in five program areas. 8

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

Note: Improper payment estimates displayed in the figure include both improper and unknown
payments. Executive agency estimates of improper payments must treat as improper any payments

5GAO, 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).
6GAO, Improper Payments: Information on Agencies’ Fiscal Year 2024 Estimates,
GAO-25-107753 (Washington, D.C.: Mar. 11, 2025).
7GAO, Program Integrity: Agencies and Congress Can Take Actions to Better Manage
Improper Payments and Fraud Risks, GAO-25-108172 (Washington, D.C.: Mar. 11, 2025).
8GAO-25-107753.

Page 3 GAO-26-108850
whose propriety cannot be determined due to lacking or insufficient documentation. 31 U.S.C §.
3352(c)(2)(A). Fiscal year 2024 data were the most recently available data on paymentaccuracy.gov.

Over the years, we have recommended ways that agencies can reduce
fraud and improper payments, and ways that Congress can help. For
example, from July 2015 through August 2023, we made 47
recommendations to federal agencies related to designing and
implementing data-analytics activities to prevent and detect fraud, such
as using data matching to verify self-reported information. One of these
recommendations—to the Small Business Administration—prompted the
agency to screen all Paycheck Protection Program loans made before
September 2020 with a rules-based data analytic tool—just looking for
indicators, not using AI. In doing so, the Small Business Administration
identified $4.7 billion in loans that went to ineligible recipients or were
used for unauthorized purposes. 9

AI presents both opportunities and challenges for combatting fraud and
improper payments in the federal government. My statement today
discusses: (1) examples of tools and resources Congress and federal
agencies can leverage to improve existing efforts to combat fraud and
improper payments without the use of AI, (2) opportunities and challenges
in using data analytics and AI to combat fraud and improper payments,
and (3) challenges in developing an AI-ready federal workforce to deploy
innovative tools to combat fraud and improper payments.

This statement is based on a body of work from July 2015 to December
2025 addressing improper payments, fraud, and artificial intelligence.
More detailed information on the scope and methodology of our prior
work can be found within the specific reports on which this statement is
based. These reports are listed in the related products page at the end of
this statement. In addition, appendix I includes a list of our selected
completed and ongoing work on AI.

We conducted the work on which this statement is based in accordance
with all sections of GAO’s Quality Assurance Framework that are relevant
to our objectives.

9GAO, COVID Relief: Fraud Schemes and Indicators in SBA Pandemic Programs,
GAO-23-105331 (Washington, D.C.: May 18, 2023).

Page 4 GAO-26-108850
While eliminating all fraud is not a realistic goal, the federal government
Congress and has tools and resources in place to help agencies prevent fraudulent and
Federal Agencies improper payments before funds are distributed and to combat fraud and
improper payments after they have occurred. Examples of these tools
Could Strengthen and resources include the Department of the Treasury’s Do Not Pay
Use of Existing Tools system, GAO’s A Framework for Managing Fraud Risks in Federal
Programs (Fraud Risk Framework), and GAO’s Antifraud Resource.
and Resources to Congressional and agency action on our matters and recommendations
Combat Fraud and in these areas could improve these efforts.
Improper Payments Department of the Treasury’s Do Not Pay system. One key resource
for preventing fraud and improper payments is the Department of the
Treasury’s Do Not Pay system, which consolidates data on entities
ineligible to receive payments. Do Not Pay is a data-matching service for
agencies to use in preventing payments to ineligible individuals, such as
those who are deceased. The Social Security Administration currently
shares its full death data with Do Not Pay, but its requirement to do so
expires in December 2026. In March 2022, we raised a matter for
Congress to consider making this requirement permanent, which would
enhance agencies’ ability to verify identities and identify payments to the
deceased. 10

The sharing of these data has already yielded positive financial results.
According to Treasury, Do Not Pay’s first year of temporary access to the
Social Security Administration’s full death data resulted in total net
benefits of $109 million, with benefits of at least $330 million projected
over the 3-year approved access period outlined in legislation (December
2023 through December 2026). 11

10We recommended that Congress amend the Social Security Act to accelerate and make
permanent the requirement for Social Security Administration 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). Bills Ending Improper
Payments to Deceased People Act, H.R. 2716, 119th Cong. (2025); S. 269, 119th Cong.
(2025), and PIIA Reform Act, H.R.1533, 119th Cong. (2026), address this.
11GAO, Confronting High Risks and Fiscal Challenges Is Crucial to a More Effective and
Sustainable Government, GAO-26-108824 (Washington, D.C.: Dec. 16, 2025).

Page 5 GAO-26-108850
Example: Matching data to identify Our prior work has also demonstrated how data analytics can be used to
potential fraud confirm the validity of a Social Security number and whether a beneficiary
The Patient Protection and Affordable Care is alive. Recently, we reported on preliminary results that showed how
Act provides premium tax credits to eligible
individuals who purchase private health
analytics involving the Social Security Administration’s death data can
insurance plans through health insurance identify potential health insurance fraud (see sidebar).
marketplaces established under the law.
Individuals may have the federal government
pay this credit to their health insurance GAO’s Fraud Risk Framework. For fraud risk management, preventive
issuers in advance on their behalf, known as activities generally offer the most cost-efficient use of resources, since
the advance premium tax credit, which lowers
the individuals’ monthly premium payments. In they allow the federal government to avoid a costly and inefficient “pay-
matching tax credit and Social Security and-chase” approach. Managers of federal programs maintain the
Administration’s death data, our preliminary primary responsibility for managing fraud risk. Since 2016, agencies have
analysis found over 58,000 deceased
individuals who had received the tax credit been required to adhere to leading practices in our Fraud Risk
(about 0.4 percent of tax credit recipients in Framework to effectively design, implement, and operate an internal
plan year 2023). These include 7,000
instances where the reported date of death control system that addresses fraud risks. 12 One leading practice in the
occurred prior to enrollment. Fraud Risk Framework involves designing and implementing data analytic
Source: GAO, Patient Protection and Affordable Care Act:
Preliminary Results from Ongoing Review Suggest Fraud controls to prevent and detect fraud (see fig. 2).
Risks in the Advance Premium Tax Credit Persist,
GAO-26-108742, (Washington, D.C.: Dec. 3, 2025). |
GAO-26-108850

Figure 2: Overview of GAO’s Fraud Risk Framework

12GAO, A Framework for Managing Fraud Risks in Federal Programs, GAO-15-593SP
(Washington, D.C.: July 28, 2015). OMB’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.

Page 6 GAO-26-108850
We have made several recommendations to federal agencies from our
body of work on fraud risk management. More specifically, we have
highlighted areas in which federal agencies need to take additional
actions to help ensure the agencies are effectively managing fraud risks
consistent with leading practices, such as using data analytics to better
manage fraud risk. For example, in February 2024 we recommended that
the Department of Defense revise its Fraud Risk Management Strategy to
establish data analytics as a method for preventing, detecting, and
responding to fraud. 13 Contrary to leading practices, we found that the
department’s strategy did not establish data analytics as a method for
fraud risk management. Data analytics can include a variety of
techniques, such as data matching to determine eligibility to receive
federal contracts. For example, we reported that if an entity reports that it
is a small business to receive federal contracts, the Department of
Defense can use third-party data sources to verify that the entity qualifies
as a small business. The Department of Defense did not agree with our
recommendation but issued its revised Fraud Risk Management Strategy
in June 2025. While the revised strategy requires components to develop
action plans for implementing data-analytics activities in certain areas by
a certain time, it does not provide specificity regarding the method for
using data-analytics activities in preventing, detecting, and responding to
fraud, nor does it discuss what or how data analytics are to be used.

Similarly, in August 2023, we recommended that the Department of
Housing and Urban Development Assistant Secretary for Community
Planning and Development identify ways to collect and combine
contractor and subcontractor data across grantees and subrecipients to
facilitate risk analyses, among other things. 14 We found that incomplete
and inconsistent data impacted grantees’ and subrecipients’ ability to
make eligibility determinations and limited the department’s ability to
oversee grantees’ processes for managing fraud risk. The department is
currently developing a Fraud Risk Playbook, which the department told us
will inform the development of guidance to implement this
recommendation. The Fraud Risk Playbook, along with tools and
templates, is intended to help program offices design and implement
fraud risk management programs, including fraud risk assessments.

13GAO, DOD Fraud Risk Management: Enhanced Data Analytics Can Help Manage
Fraud Risks, GAO-24-105358 (Washington, D.C.: Feb. 27, 2024).
14GAO, Disaster Recovery: HUD Should Develop Data Collection Guidance to Support
Analysis of Block Grant Fraud Risks, GAO-23-104382 (Washington, D.C.; Aug. 17, 2023).

Page 7 GAO-26-108850
Antifraud Resource. To further help agencies build prevention-focused
antifraud efforts, in 2022, we developed a web-based Antifraud Resource,
which provides interactive tools and resources for understanding and
combatting fraud, such as fraud scheme examples (see fig. 3). 15

Figure 3: Five Categories of Antifraud Resources

15GAO, “The GAO Antifraud Resource” (Washington, D.C.: Jan. 10, 2022),
https://gaoinnovations.gov/antifraud_resource/.

Page 8 GAO-26-108850
Opportunities and
Challenges in Using
Data Analytics and AI
to Combat Fraud and
Improper Payments
Routine and Innovative As we have previously reported, there are a range of data analytic tools—
Data Analytic Tools Can from basic to advanced—that can help prevent and detect fraudulent
activity and improper payments in federal programs (see fig. 4). The
Help Reduce Fraud and
advanced tools can be particularly helpful in identifying scale,
Improper Payments connections, and outliers.

Page 9 GAO-26-108850
Figure 4: Examples of Data Analytic Tools for Preventing and Detecting Fraud in
Federal Programs

Note: More information about how data analytic tools can be used to prevent and detect fraud is
available in GAO, A Framework for Managing Fraud Risks in Federal Programs, GAO-15-593SP
(Washington, D.C.: July 28, 2015); and Chief Financial Officers Council and the Bureau of the Fiscal
Service, Program Integrity: The Antifraud Playbook (Oct. 17, 2018).

Page 10 GAO-26-108850
Using AI to Reduce Fraud As we have previously reported, new innovations like AI-enabled tools
and Improper Payments can perform some of the tasks required to combat fraud and improper
payments in the federal government. 16 However, these tools require
Requires High-Quality
sufficient high-quality data, an understanding of risks, and staff with
Data and Responsible Use knowledge of how to use AI. Introducing insufficient, unrelated, or bad
data will make an AI model less reliable, and maintaining a “human in the
Indicators of fraudulent activity from loop” is vital to ensuring oversight of the data and processes. 17 An AI-
organized groups enabled tool that produces errors will also erode trust in AI’s ability to
According to federal and state officials, when
fraudsters work together in groups, data
detect fraud—and therefore, the government’s as well.
analytics can identify the following indicators
of their fraudulent activity: AI, in general, refers to computer systems that can solve problems and
Scale. A large number of actions occurring perform tasks that have traditionally required human intelligence. There
within a short period of time. For example,
groups of fraudsters may submit thousands of are several subcategories of AI such as machine learning and generative
applications for public benefits within a short AI (see fig. 5). Machine learning underpins many of the recent
period of time that may use similar identifying
information. improvements in AI and could be used to detect fraud and improper
Connections. Relationships between people, payments. Generative AI refers to tools that can create text, images,
entities, or other associated data points. For audio, and other content using patterns learned from datasets.
example, multiple participants connected by
the same personally identifiable information
may be used to obtain information on
thousands of Supplemental Nutrition
Assistance Program Electronic Benefit
Transfer cards.
Outliers. Data points outside of what would
ordinarily be seen. For example, an organized
fraud group might submit reimbursement
claims to Medicaid that are unreasonable
based on geography or time.
Source: GAO, Fraud Risk in Federal Programs: Continuing
Threat from Organized Groups Since COVID-19,
GAO-25-107508 (Washington, D.C.: July 10, 2025). |
GAO-26-108850

16GAO-25-108412.

17Human in the loop refers to an approach to human oversight of the AI system “with the
human retaining full control and the AI only providing recommendations or input.” A
human reviews the output of the AI system and makes the final decision. Info-
communications Media Development Authority and Personal Data Protection
Commission, Model Artificial Intelligence Governance Framework, Second Edition
(Singapore, Jan. 2020).

Page 11 GAO-26-108850
Figure 5: Generative Artificial Intelligence (AI) and Machine Learning in Relation to Other Types of AI

Page 12 GAO-26-108850
Many federal agencies are using or planning to use machine learning,
including applications to support detecting fraud and improper payments.
For example, in November 2020, the Small Business Administration
worked with its loan-review contractor to put in place a machine-learning
tool to rate borrower-forgiveness applications according to fraud risk and
to clear batches of loans flagged during automated screening that were
considered low risk. 18

Machine-learning models work by identifying statistical relationships
between inputs and outputs from a training dataset. Training is the
process of feeding training data through the algorithm until it identifies
statistical relationships of interest. Such models 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. 19

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.

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. For detecting fraud and
improper payments, incorrect labeling in historical data in the model could
lead to false results. If these become too numerous, agencies could
spend more time identifying AI’s mistakes than they would 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.

To help address these potential challenges, we have reported that
agencies need solid, reliable “ground truth” data and a human in the loop
to ensure data reliability and appropriate application of the technology. 20
AI does not replace the professional judgment of experienced staff in

18GAO, COVID-19 Relief: Improved Controls Needed for Referring Likely Fraud in SBA’s
Pandemic Loan Programs, GAO-25-107267 (Washington, D.C.: Mar. 24, 2025).
19Taka Ariga, “Artificial Intelligence Creates New Opportunities to Combat Fraud,”
International Journal of Government Auditing, Summer 2020 Edition.
20GAO, Artificial Intelligence: An Accountability Framework for Federal Agencies and
Other Entities, GAO-21-519SP (Washington, D.C.: June 30, 2021).

Page 13 GAO-26-108850
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.

In addition, federal government data for identifying possible fraud and
improper payments can vary in quality, reliability, and availability. For
example, in our 2023 survey on fraud risk management, federal agencies
told us that access to data to look for fraud indicators was a challenge. 21
We have also previously reported on challenges with federal agency data
used to identify possible fraud and improper payments at the Federal
Aviation Administration and Small Business Administration, among
others. 22 We have made recommendations to federal agencies to help
improve their use of data, such as recommending agencies implement
data mining and other analytic practices to ensure data are sufficient for
fraud risk analysis.

Moreover, to improve the use of data analytics in identifying fraud and
improper payments, in 2022 we recommended that Congress establish a
permanent analytics center of excellence. 23 We noted that such a center
could include an AI-based tool to help in identifying fraud and improper
payments. In July 2025, Congress extended the sunset date of the
Pandemic Response Accountability Committee to September 30, 2034,
and expanded its jurisdiction. The committee includes the Pandemic
Analytics Center of Excellence; however, the committee and its analytics
center are not permanent or government-wide.

In addition, while private-sector enterprises are exploring using generative
AI to prevent and detect fraudulent activity, we have found that generative
AI has challenges securing classified and sensitive data that private-
sector users may not face. 24 In July 2025 we reported that, according to
Department of Defense officials, generative AI models could aggregate

21GAO, Fraud Risk Management: Agencies Should Continue Efforts to Implement Leading
Practices, GAO-24-106565 (Washington, D.C.: Nov. 01, 2023).
22GAO, Aviation: FAA Needs to Better Prevent, Detect, and Respond to Fraud and Abuse
Risks in Aircraft Registration, GAO-20-164 (Washington, D.C.: March 25, 2020) and GAO,
Small Business Research Programs: Opportunities Exist for SBA and Agencies to Reduce
Vulnerabilities to Fraud, Waste, and Abuse, GAO-24-105470 (Washington, D.C.: Sept. 9,
2024).
23GAO-22-105715.

24GAO, Artificial Intelligence: Generative AI Use and Management at Federal Agencies,
GAO-25-107653 (Washington, D.C.: July 29, 2025).

Page 14 GAO-26-108850
various unclassified information contained in the model’s training data
and unintentionally output classified information. We reported that federal
agencies are required to ensure that personal data, controlled
unclassified information, and classified data used in the training and
deploying of generative AI models are kept secure and compliant with
federal requirements. Officials at five agencies told us that strict data
security requirements may prevent them from performing generative AI
research in certain agency mission areas.

In recognition of these opportunities and risks of AI, in 2021, we published
our AI Accountability Framework for federal agencies and other entities
(see fig. 6). 25 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.

25GAO-21-519SP.

Page 15 GAO-26-108850
Figure 6: GAO Artificial Intelligence (AI) Accountability Framework

Page 16 GAO-26-108850
An AI-ready workforce is important for the federal government in using AI
Challenges in to combat fraud and improper payments in federal programs. For
Developing an AI- decades, however, we have identified mission-critical gaps in science,
technology, engineering, and mathematics skills and expertise within the
Ready Federal federal workforce. In December 2023 and July 2025, we reported on
Workforce to Combat challenges in attracting and developing individuals with AI expertise in
federal agencies, including Departments of Commerce, Defense, State,
Fraud and Improper and Veterans Affairs; the National Aeronautics and Space Administration;
Payments and the Office of Personnel Management. 26 These challenges could also
apply to creating a workforce that can use AI to prevent and detect fraud.

One option Congress has considered is to establish a new digital services
academy—like the military academies—to train future workers. 27 In 2021,
we convened technology leaders from government, academia, and
nonprofits to discuss such an academy and related issues. They raised
several comments including that:

• Current federal digital staff compensation is not competitive.
• 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 Sessions, Ranking Member Mfume, and Members of the
Subcommittee, this completes my prepared statement. I would be
pleased to respond to any questions that you may have at this time.

Sterling Thomas, ThomasS2@gao.gov
GAO Contact and
Rebecca Shea, SheaR@gao.gov
Staff
Acknowledgements In addition to the contacts named above, Joseph Cook (Assistant
Director), Heather J. Dunahoo (Assistant Director), Claire McLellan
(Analyst in Charge), Lisa Gardner, Allison Henn, Riley Wilson, Anika
McMillon, and Victoria Aysola made key contributions to this testimony.
Other staff who contributed to this testimony include Daniel Flavin,
Rachael Johnson, and Jessica Steele.

26GAO, Artificial Intelligence: Actions Needed to Improve DOD’s Workforce Management,
GAO-24-105645 (Washington, D.C.: Dec. 14, 2023); GAO-25-107653.
27GAO, Digital Services: Considerations for a Federal Academy to Develop a Pipeline of
Digital Staff, GAO-22-105388 (Washington, D.C.: Nov. 19, 2021).

Page 17 GAO-26-108850
Appendix I: A Selection of Artificial
Appendix I: A Selection of Artificial Intelligence
Work from GAO

Intelligence Work from GAO

Page 18 GAO-26-108850
Appendix I: A Selection of Artificial Intelligence
Work from GAO

Page 19 GAO-26-108850
Related GAO Products
Related GAO Products

Confronting High Risks and Fiscal Challenges Is Crucial to a More
Effective and Sustainable Government. GAO-26-108824. Washington,
D.C.: December 16, 2025.

Artificial Intelligence: Generative AI Use and Management at Federal
Agencies. GAO-25-107653. Washington, D.C.: July 29, 2025.

Fraud Risk in Federal Programs: Continuing Threat from Organized
Groups Since COVID-19. GAO-25-107508. Washington, D.C.: July 10,
2025.

Fraud and Improper Payments: Data Quality and a Skilled Workforce Are
Essential for Unlocking the Benefits of Artificial Intelligence.
GAO-25-108412. Washington, D.C.: April 9, 2025.

COVID-19 Relief: Improved Controls Needed for Referring Likely Fraud in
SBA’s Pandemic Loan Programs. GAO-25-107267. Washington, D.C.:
March 24, 2025.

Program Integrity: Agencies and Congress Can Take Actions to Better
Manage Improper Payments and Fraud Risks. GAO-25-108172.
Washington, D.C.: March 11, 2025.

Improper Payments: Agency Reporting of Payment Integrity Information.
GAO-25-107552. Washington, D.C.: January 23, 2025.

Small Business Research Programs: Opportunities Exist for SBA and
Agencies to Reduce Vulnerabilities to Fraud, Waste, and Abuse.
GAO-24-105470. Washington, D.C.: September 9, 2024.

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.: April
16, 2024.

Artificial Intelligence: Actions Needed to Improve DOD’s Workforce
Management. GAO-24-105645. Washington, D.C.: December 14, 2023.

Improper Payments and Fraud: How They Are Related but Different.
GAO-24-106608. Washington, D.C.: December 7, 2023.

Page 20 GAO-26-108850
Related GAO Products

Fraud Risk Management: Agencies Should Continue Efforts to Implement
Leading Practices. GAO-24-106565. Washington, D.C.: November 01,
2023.

Disaster Recovery: HUD Should Develop Data Collection Guidance to
Support Analysis of Block Grant Fraud Risks. GAO-23-104382.
Washington, D.C.: August 17, 2023.

COVID Relief: Fraud Schemes and Indicators in SBA Pandemic
Programs, GAO-23-105331. Washington, D.C.: May 18, 2023.

Digital Services: Considerations for a Federal Academy to Develop a
Pipeline of Digital Staff. GAO-22-105388. Washington, D.C.: November
19, 2021.

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

Aviation: FAA Needs to Better Prevent, Detect, and Respond to Fraud
and Abuse Risks in Aircraft Registration. GAO-20-164. Washington, D.C.:
March 25, 2020.

A Framework for Managing Fraud Risks in Federal Programs.
GAO-15-593SP. Washington, D.C.: July 28, 2015.

Page 21 GAO-26-108850
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

David A. Powner, Acting 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-26-108850.pdf
Size: 2,904,627 bytes
SHA-256: 32bd8516ec04714e20f9ee3df7967517156c4453464aa2ab9ca76226bf797e7e
Our copy: gao-26-108850.pdf
Original: www.gao.gov