Public money carries a promise. When citizens pay taxes or businesses clear goods through customs, they trust the government to collect funds fairly, safeguard them diligently, and use them for the public good. When a family receives a benefit payment, that promise runs in reverse: the right support reaches the right person at the right time. Protecting public money means protecting that promise, and with it, public confidence in government.
Today, that promise is under unprecedented strain. Financial crime has become faster, more organized, and more automated, while public finance leaders are asked to do more with less. At the same time, advances in agentic AI, Zero Trust security, and secure, interoperable data platforms can help tax, treasury, customs, and benefits agencies shift from reacting to fraud after the fact to preventing and detecting it in real time.
The industry context: Rising risk, rising expectations
The scale of the challenge has moved from serious to staggering. In the United States, the Government Accountability Office now estimates that the federal government could lose between $233 billion and $521 billion to fraud every year.1 Improper payments totaled an estimated $187 billion in FY 2025.2 Globally, the United Nations Office on Drugs and Crime estimates money laundering at 2% to 5% of world gross domestic product (GDP), on the order of $800 billion to $2 trillion annually.3
The threat is also changing. In the anti-financial-crime community’s 2025 global survey, 75% of anti-financial-crime professionals rated malicious generative AI (such as deepfakes, synthetic identities, and automated scams) as a high or very high external risk.4 Fraud can now be launched faster, scaled more cheaply, and made harder to detect with traditional rules-based checks alone.
However, not all loss stems from criminal intent. A significant share comes from unintentional error: administrative mistakes, poorly designed processes, or a citizen submitting incorrect information. Public finance organizations must address both fraud and error while operating under pressures that private-sector firms rarely face:
- Rising demand and complexity. Aging populations and broader entitlements are driving larger budgets and more intricate programs.
- Mandates to do more with less. Austerity and budget pressure collide with rising public expectations for modern, digital services.
- Disconnected data silos. Departments often run isolated systems that cannot share information easily or securely.
- Legacy technical debt. Outdated systems slow innovation, increase operating risk, and can create delays or errors that prevent eligible people from receiving timely support.
The opportunity is equally striking, and the economics decisively favor prevention. When the United States Treasury strengthened its payment checks against a central “do-not-pay” data source, its first-year pilot results reported preventing or recovering $113.5 million against a $4.6 million cost, roughly a 23-to-1 return.5 Yet adoption remained limited, with fewer than one in ten federal agencies use such payment-integrity tools. Modern cloud and AI capabilities can help close that gap: simplifying revenue collection, directing resources more precisely, and combating fraud and corruption while making services faster and fairer for legitimate users.
Paradigm shifts: Agentic AI, Zero Trust, and security by design
Three converging shifts are changing what is possible, and each one matters most precisely when trust is the currency at stake.
1. From copilots to agents: The rise of agentic AI
The first shift is from AI that assists to AI that can act within defined boundaries. Public-sector organizations are progressing from assistants that help people find and summarize information toward agentic systems that can complete governed, multi-step tasks.
Agents could help financial-crime teams monitor transactions, triage alerts, gather case evidence, summarize findings, and route higher-risk cases to human investigators. With strong governance, clear accountability, and meaningful human oversight, these capabilities can help transform reactive, sample-based processes into more proactive, continuous ones.
2. Zero Trust: Never trust, always verify
Public finance organizations hold some of the most sensitive data any institution manages, including taxpayer records, financial histories, and identity information. A Zero Trust approach verifies explicitly, applies least-privilege access, and assumes that breaches can occur. This security foundation can help agencies use data for AI-powered analysis and cross-agency collaboration while helping reduce the risk of unauthorized access or misuse.
3. Security and responsibility by design
Trust cannot be added after deployment. Agencies need AI systems that are transparent, controllable, secure, and accountable by design.
Microsoft’s approach is guided by its Responsible AI Standard and includes capabilities such as content provenance and red-teaming. Just as important is meaningful human oversight. In high-stakes public finance decisions, AI should support human judgment, not replace it. Decisions that affect a person’s taxes, benefits, or legal rights should remain explainable and auditable, with accountability resting with an authorized person.
Building blocks for financial crime prevention and detection
Turning these shifts into results calls for an integrated architecture rather than a single tool. Microsoft unifies these building blocks into a secure, coherent whole that can help revenue and finance agencies detect and prevent tax fraud and financial crime:
- A unified, secure data foundation. Breaking down silos so data can be securely brought together and shared across organizations is the prerequisite for everything else. Equally important is preserving data sovereignty, enabling agencies to maintain control over where sensitive information is stored, who can access it, and how it is used. Confidential computing and privacy-preserving techniques can support collaboration while helping protect sensitive information.
- Trusted digital identity. Verifiable credentials and (increasingly) quantum-safe digital identity can help reduce synthetic-identity and account-takeover fraud at the point of entry.
- Advanced analytics and anomaly detection. A 360-degree view of the taxpayer or beneficiary, combined with AI-powered risk profiling across the value chain, can surface anomalous patterns associated with potential fraud, waste, or error.
- Network and entity intelligence. Contextual decision-intelligence connects internal records with external sources that may help reveal the hidden networks and relationships behind potential organized fraud, not just isolated bad transactions.
- Agentic automation. AI agents can help scale scarce investigative capacity: monitoring accounts, correlating signals, drafting case summaries, and accelerating audits so specialists can focus on the highest-value work.
- Zero Trust security and cyber-resilience. Identity-centric access, continuous verification, and layered protection safeguard the platform and the data that flows through it.
- Responsible AI guardrails. Transparency, auditability, and human-in-the-loop review are engineered in, so that detection is not only powerful but also fair, contestable, and accountable.
Together, these building blocks can help agencies detect fraud and prevent improper payments earlier, while making legitimate services faster and simpler for citizens and businesses.
Why the government ecosystem matters
Financial crime does not respect organizational charts or national borders, and neither can the response. Protecting public money is a team effort. No single agency can win it alone.
On the public side are tax and revenue authorities, treasuries and finance ministries, customs and border agencies, social-benefit and welfare bodies, financial intelligence units, supreme audit institutions, public-sector fraud authorities, national digital-identity providers, and law enforcement. On the private side are banks and payment processors, fintechs and digital wallets, telecom operators, and identity-verification and technology partners, each of which sees signals that government may not. Evidence of a stolen identity used by a fraud ring may surface first at a bank; a duplicate benefit claim only at an agency; a mule account only within a payment network. Fraud becomes visible when those fragments are brought together.
A longstanding barrier to that cooperation has been trust. Institutions cannot simply pool sensitive citizen and financial data without introducing risks to privacy, security, and sovereignty. Modern technology can help overcome that barrier. Confidential computing and privacy-preserving analytics are designed to help organizations identify matches and patterns while minimizing exposure of underlying records. Secure data-collaboration environments—including data clean rooms and “data trust zones”—enable partners to conduct joint analysis under strict, auditable governance. Real-time data services stream signals as they appear, while AI agents correlate them across sources, helping turn scattered fragments into potentially meaningful insights quickly. Verifiable digital identity and a Zero Trust foundation help keep every exchange governed, least-privilege, and fully auditable.
Proof in practice: Agencies fighting fraud with AI
These are not future ambitions. Across regions, public finance, and benefits organizations are already using Microsoft AI, including emerging agentic capabilities, to protect public money.
United States—Conduent
Conduent, which disburses roughly $85 billion in government payments each year and supports public programs across 37 U.S. states, completed a generative AI pilot with Microsoft (now fully deployed), that expanded its ability to review potential fraud signals. Because its open-loop payment cards can be used at a wide range of merchants, monitoring is especially complex. With AI, a small team of specialists can now monitor tens of thousands of accounts for suspicious activity, including identity theft and account takeover, with reported improvements in its review processes. The capability is now being scaled to additional programs, including Medicaid and SNAP benefits.6
United Kingdom—HM Revenue and Customs
The United Kingdom’s tax authority has expanded its partnership with Microsoft to deploy generative and agentic AI across its operations. It has rolled out Microsoft 365 Copilot to 28,000 employees, with plans to reach 50,000. AI agents summarize customer complaints and help advisers quickly find the information they need to resolve cases. Crucially, HMRC has been explicit that AI will not make final determinations on tax cases: “Any decisions will be reviewed and signed off by a human in the process with the relevant tax expertise.” The department has also appointed its first Chief AI Officer to lead the transformation.
Around the world
The pattern repeats across regions. In the Middle East and Africa, the Egypt Tax Authority built an e-invoicing and e-receipt system on the Microsoft Cloud to tackle a costly shadow economy, cutting tax examinations from arbitrary estimates to assessments completed in hours and laying the groundwork for early-stage fraud detection. In the Americas, Microsoft worked with the Inter-American Center of Tax Administrations, which supports 42 member countries, to build intelligence tools that detect anomalies and help tax agencies work more efficiently; and the General Directorate of Customs of the Dominican Republic modernized operations on the Microsoft Cloud, cutting some processes from a week to under 24 hours while strengthening border risk controls.
Take the next step
Protecting public money, strengthening accountability, and earning public trust are among government’s most consequential missions. In the age of agentic AI, advances in data, security, analytics, and automation give agencies new ways to pursue them. Early adopters are showing that governments can strengthen their ability to identify potential financial crime earlier and respond more effectively, while making services simpler and fairer for the citizens they serve.
A key Microsoft strength is the ability to bring these capabilities together in a unified cloud and AI platform: a secure data foundation, Zero Trust security, agentic AI, and responsible AI governance.
To explore the strategies, safeguards, and practical roadmap behind this transformation, read the Microsoft e-book Public Finance Management in the Age of AI, a guide for public finance leaders ready to put these ideas to work.
1 U.S. Government Accountability Office, Fraud Risk Management, 2024.
2 “Fraud & Improper Payments,” U.S. Government Accountability Office, July 2026.
3 United Nations Office on Drugs and Crime (UNODC), Money Laundering: Overview.
4 ACAMS, Global AFC Threats Report 2025.
5 U.S. Government Accountability Office, Social Security Death Data, 2026.
