Do these 3 things before closing this tab:
1Clear out junk files and repair common Windows errors2Fix the driver behind crashes, sound loss and screen glitches3Repair Windows errors before they cause bigger problemsiTechGuides is reader-supported. When you buy through links on our site, we may earn an affiliate commission. As an Amazon Associate I earn from qualifying purchases. Learn more
AI is being used in financial services and public agencies to review documents, flag suspicious transactions, propose fraud-detection rules, inspect identity images, and support market surveillance. These systems do not all make decisions in the same way: some surface cases for human review, while others analyze broad streams of signals or intervene in a customer payment journey. Their reported results also vary in evidential strength, from an independently audited public-sector deployment to company announcements and anonymized case studies.
Where AI is being used
1. Medicare provider fraud detection in Australia
Australia’s Department of Health, Disability and Ageing began using an AI-enabled model in its Medicare fraud detection system in July 2024. In December 2025, it replaced that model with non-AI logistic regression to reduce the number of providers flagged. The Australian National Audit Office’s audit found gaps in recorded output validation and business approval, limited documentation of deployment testing, no implemented fairness metrics, and informal, irregular production monitoring. The audit also describes planned improvements to governance and monitoring. This is a useful reminder that a model can be changed when its operational results are unsuitable, and that oversight needs to cover development, approval, deployment, monitoring, and withdrawal—not just the initial launch.
2. Sales-quality compliance at an unnamed UK bank
A UK Government case study published in 2019 describes a large global bank that used AI to review financial-product sales for regulatory compliance. Before automation, 120 reviewers sampled 10% to 15% of completed sales. Each review took around four hours and drew on more than 10 data sources and 180 data points. The materials included structured information as well as letters, memos, payslips, and bank-statement images; the case study says at least 70% of checks involved unstructured data.
The bank and its technology provider tested models against real data and reviewer feedback before live deployment. The case study reports that the system enabled review of all cases, eliminated the backlog, and brought checks closer to real time. It also reports close to 100% accuracy for automated checks. Those are the case study’s claims, not an independent benchmark; they should not be read as a general accuracy rate for AI compliance tools.
#1 Best Overall
3. Transaction monitoring at Bunq
A 2023 UK Government assurance case study describes Bunq’s transaction-monitoring workflow and its focus on making model outputs explainable. An analyst checks every transaction the model flags. The analyst either clears the flag or approves escalation to the financial intelligence unit, recording why a cleared alert was a false positive. Experts also review a small sample of unusual transactions that were not flagged, a check intended to help find false negatives.
The case study reports that explainability improvements reduced time spent on false-positive cases by about 80% and time per case by almost 90%. These are case-study-reported outcomes for this workflow, not independent measurements or estimates of what other institutions should expect.
4. Credit scoring and fraud detection in an ECB bank sample
European Central Bank Banking Supervision’s 2025 account draws on supervisory reporting for 107 significant institutions in 2023 and 110 in 2024, and workshops with 13 banks using AI in relevant cases. It describes applications including fraud detection and credit scoring, alongside governance practices such as explainability tools, centralized dashboards, human validation for higher-risk decisions, and feedback loops. Banks also raised concerns about data quality and third-party risk. None of the banks in the workshop sample permitted systems to keep self-learning after deployment.
Free tools Windows power users keep installed
One-click scans. No signup required.
Rank #2
The ECB explicitly cautions that workshop findings come from a small sample and should not be generalized to the whole sector. The counts of institutions in supervisory reporting are not counts of banks using any particular fraud or credit system. The ECB also notes that whether logistic regression qualifies as AI under the EU AI Act is not clear in the context it discusses, so the label “AI” does not identify one consistent model class.
5. Agentic fraud-rule proposals at an unnamed international bank
The Financial Stability Board’s 2026 consultation report describes an anonymized, internationally active bank that built an agentic system to monitor signals, assess suspicious patterns, and propose new fraud-detection rules. Human reviewers in the fraud analytics team approve every new rule before it goes live. In this example, the bank’s existing AI capabilities monitored more than 80 million signals each day; the report says the agent contributed to developing or updating three quarters of card-fraud rules. It says the system was built in three months and that fraud losses fell by over 20% in the first half of financial year 2026 compared with the same period in 2025.
These figures are reported for an unnamed bank in the FSB’s illustrative case study. The report does not identify the institution or establish the result through independent verification, so the loss reduction should not be treated as a forecast or a result typical of agentic AI.
Rank #3
6. Scam Check in Lloyds Banking Group payment journeys
Lloyds Banking Group announced Scam Check for certain payment journeys across Lloyds, Halifax, and Bank of Scotland. In the described journey, a customer trying to pay a new recipient for an online purchase could be asked questions and prompted to upload screenshots of the item if the payment appears risky. The announcement describes Envoy as the group’s secure platform for deploying AI agents with oversight and accountability.
Recommended Free Tools
The announcement described a forthcoming tool and did not report measured fraud-prevention results. Its rollout status and availability may have changed since the announcement, so the announcement alone does not establish what customers can use now. A statement in the release by Tom Martin, Business Platform Lead for Economic Crime Prevention at Lloyds Banking Group, framed the approach as applying agentic AI with human oversight; that is the company’s view, not evidence of effectiveness.
7. Facial-image checks at an unnamed digital bank
The FSB describes a digital bank that extended a facial-recognition solution, originally used to compare customer images, to look for suspicious image backgrounds associated with mule accounts or fraudulent identity attempts. The report does not name the bank. It presents one example, not an established standard for identity verification or a guarantee that background-image analysis identifies fraud reliably.
Rank #4
8. Marketing and disclosure review at asset managers
The FSB says some large asset-management companies use AI to support compliance review of marketing and disclosure documents, with the aim of improving the speed, consistency, and quality of checks. This is a category-level description, not a named institution’s case study, and the report does not provide a specific measured outcome for it.
9. Market surveillance and abuse detection
The FSB says some large financial market infrastructures, notably exchanges, use AI to analyze internal and external data for market surveillance and abuse detection. This, too, is a category-level account rather than a named deployment. The report also describes machine-learning fraud prevention in payment transactions and generative-AI support for AML investigations as broader application areas; those categories are additional context, not extra named cases in the nine examples above.
What makes these systems governable
A fraud or compliance model is not governed merely because a person is somewhere in the workflow. Governance depends on what the system is allowed to do, what the reviewer can see and change, and whether the institution can reconstruct and evaluate its behavior over time.
Best Value
- Used Book in Good Condition
- Set authority boundaries. Distinguish a system that flags or prioritizes a case from one that recommends an action, proposes a rule, or directly changes a customer’s experience. For high-impact actions, specify who can approve, override, clear, or escalate an output.
- Validate before release. Test the model and its full workflow on relevant data, document the intended use and approval, and define what evidence is required before production. The Australian audit shows why testing and business approval need durable records.
- Make decisions traceable. Preserve relevant inputs, model versions, outputs, explanations, reviewer actions, and rule changes so that an alert or decision can be examined later. Explainability is useful only when it helps a reviewer understand and act on an output.
- Monitor after launch. Set regular checks for changing data, performance, false positives, and false negatives. Review unusual cases that the system did not flag as well as alerts it raised; track who responds when the system drifts or produces an unsuitable volume of flags.
- Control updates and third parties. Record model and configuration changes, keep approval gates for consequential changes, and assess external models and providers rather than assuming a vendor’s assurances are sufficient. A 2023 UK Government case study on third-party model due diligence involving Deloitte illustrates that assurance itself has limits: it reported that provider processes and controls were not well documented.
- Plan for rollback or replacement. Define how the organization can restrict, change, or stop a system if its outputs or operating conditions become unsuitable. Keep the alternative process clear enough to maintain service while the system is reviewed.
How to read reported results
The numbers in these examples do not measure the same thing. The Medicare case was examined by the Australian National Audit Office, which also documented governance weaknesses. The UK bank and Bunq results come from government case studies, the ECB describes a limited supervisory and workshop sample, and the FSB’s quantified agentic example is anonymized. Lloyds’ announcement describes a planned customer journey without an effectiveness statistic. A deployment claim, a reported operational improvement, and an independently verified outcome are different kinds of evidence.
When evaluating an AI fraud or compliance claim, look for the population and period covered, the outcome being measured, the method of evaluation, and whether the source is independent of the deploying organization. Also ask whether the evidence covers false negatives and customer impact, not just processing speed or reduced review time.
Quick Recap
Product prices and availability are accurate as of the date/time indicated and are subject to change. Any price and availability information displayed on Amazon at the time of purchase will apply.
