How does Mastercard use AI for fraud and risk?
Mastercard applies fraud AI at four points: the card authorisation, the account-to-account transfer, the compromised card and the merchant. Decision Intelligence scores transactions at authorisation, and Decision Intelligence Pro, announced in February 2024, added generative AI; Mastercard's initial modelling showed fraud detection up 20% on average and false positives down more than 85%. In May 2024 Mastercard said generative AI had doubled its detection rate of compromised cards. On real-time transfers, Consumer Fraud Risk was used by 15 banks representing 90% of UK account-to-account payments by July 2025, and a February 2025 partnership with Feedzai is taking it to more markets. For merchants, Mastercard announced Merchant Trust Services and a Merchant Scam & Risk Indicator on May 19, 2026; in a pilot with one issuer the indicator detected about 80% of the issuer's risky merchants. Mastercard says its AI-enabled Safety Net has prevented $77 billion in fraud losses since 2020 and that AI secures more than 159 billion transactions a year. Nearly all results are Mastercard's own measurements; a Forrester Total Economic Impact study on Mastercard's site reports 12% lower fraud losses for users of Decision Intelligence & Rules Services. Mastercard does not report revenue from its fraud products separately.[1][2][3][4][5][6][7][8][9]
Data as of Oct 19, 2026 · Mastercard profile updated Oct 19, 2026 · 25 sources cited on this page · 52 fintechs profiled on this site · Markdown
Which fraud and risk models does Mastercard run, and what does each do?
One row per model or product, with the latest result Mastercard has published for it. Results are Mastercard's own measurements unless stated.
| Model or product | What it does | Reported result | Date |
|---|---|---|---|
| Decision Intelligence Pro | Generative AI added to authorisation fraud scoring; scans one trillion data points and returns a score in under 50 milliseconds | Initial modelling: fraud detection up 20% on average and as high as 300% in some instances; false positives down more than 85% | February 1, 2024[1] |
| Compromised-card detection | Generative AI predicts the full details of partly published stolen card numbers so banks can block and reissue the cards | Detection rate of compromised cards doubled; false positives against potentially compromised cards cut by up to 200%; at-risk merchants identified 300% faster | May 22, 2024[2] |
| Consumer Fraud Risk | Scores account-to-account payments for scam risk, giving intelligence to both the sending and the receiving bank | Nine UK banks at launch; 15 banks representing 90% of UK account-to-account payments by July 2025; Feedzai partnership to deploy it in more markets | 2023–2025[10][3][4] |
| Brighterion | Real-time, 24/7 transaction risk scoring; analysts add rules, such as one flagging geographically impossible transactions | Extended to over 60,000 merchants through Network International | 2024–2025[11][7] |
| Merchant Scam & Risk Indicator | Gives issuers merchant risk signals during authorisation; part of Merchant Trust Services for acquirers and payment service providers | In a pilot with a leading issuer it detected about 80% of the issuer-identified risky merchants, many as early as 90 days before the issuer's first escalation; Europe and the U.S. first | May 19, 2026[5] |
| Large tabular foundation model | Trained on billions of anonymized transactions with Nvidia and Databricks; Mastercard plans to add fraud, authorization and chargeback data, and cybersecurity is the first target | Outperformed standard industry machine-learning techniques in Mastercard's testing; no figures published | March 17, 2026[12] |
| Agentic transaction probability score | Estimates the likelihood that a transaction was initiated by an AI agent, so issuers can approve legitimate agent-led purchases | Rolling out for testing in the U.S.; Mastercard works with Cloudflare on payment and web signals and with Skyfire on Know Your Agent verification | September 30, 2026[13] |
| Mastercard Threat Intelligence | Joins Recorded Future threat intelligence with Mastercard fraud insights | Six months of market testing found malicious domains that had hit nearly 9,500 e-commerce sites, linked to an estimated $120 million in fraud | October 27, 2025[14] |
What results has Mastercard published for its fraud AI?
Mastercard's own figures unless a third party is named. The figures use different bases and periods, so they do not add up to one total.
| Measure | Figure | Date |
|---|---|---|
| Fraud losses prevented by Safety Net | $77 billion since 2020, per Mastercard's 2025 impact report | July 15, 2026[6] |
| Fraud losses stopped by AI-powered cybersecurity | Over $35 billion in the three years before July 2023 | July 6, 2023[10] |
| Network attack in 2023 | Safety Net detected more than 2,300 suspicious transactions per hour over three days and blocked $14.7 million of fraud | April 25, 2025[15] |
| Payments fraud detected, Q1 2025 | AI-powered decisioning detected 40% more than in Q1 2024, per PYMNTS' report of the earnings call | May 1, 2025[16] |
| UK authorised push payment scams | Value down more than 12% since Consumer Fraud Risk went live in 2023, citing Payment Systems Regulator data; APP fraud cases down 20% in 2024, with Mastercard saying the tool played a key role | 2025[4][3] |
| Forrester Total Economic Impact study | Users of Decision Intelligence & Rules Services: 98% ROI and $595,000 net present value over three years, 12% lower fraud losses, 20% fewer fraud alerts | Undated page, retrieved October 19, 2026[8] |
| Scam-merchant ring (Get Wuffy) | 250 lookalike storefronts with one owner; $50 million in transactions on Mastercard's network tied to the merchant IDs found; the acquirer ended the relationship and listed the merchant in MATCH Pro | October 13, 2026[17] |
| Stolen card records for sale | Down almost 20% in 2025, while 10,500 Magecart skimming hacks were active and compromised over 23 million online transactions, per Recorded Future's annual report | March 2, 2026[18] |
Which rules, regulators and dispute questions shape Mastercard's fraud AI?
| Item | What the record shows | Date |
|---|---|---|
| Scam-merchant enforcement rule | From July 2026 acquirers and payment facilitators must open an investigation within 72 hours when potential scam activity hits a set risk threshold; confirmed scam merchants must be stopped from accepting Mastercard | May 19, 2026[5] |
| First-Party Trust | Data-sharing and compelling-evidence rules for 'friendly fraud' disputes, expanded from the U.S. to Canada, Latin America, the Caribbean and Asia Pacific; Mastercard forecasts chargebacks costing merchants $42 billion by 2028 | June 25, 2025[19] |
| UK Payment Systems Regulator data | Mastercard cites the regulator's APP scams performance report to support Consumer Fraud Risk's results | February 18, 2025[4] |
| EU AI Act and model risk | The FY2025 10-K says Mastercard is subject to the EU AI Act as a user and deployer of AI, and that AI algorithms may generate inaccurate, unintended, unfair, biased or discriminatory outcomes | February 11, 2026[9] |
| Know Your Agent framework | Mastercard, Visa and Ant International began work on an interoperability framework to onboard and identify AI agents across networks, with continuous monitoring of agents | September 17, 2026[20] |
| Liability for agent purchases | Mastercard told Fortune its Verifiable Intent record could help resolve a dispute, but did not specify who is responsible if an agent buys outside the cardholder's instructions | September 18, 2026[21] |
How does Mastercard run its fraud AI, and what does it not disclose?
Mastercard pairs models with investigators. Its October 2025 account says AI secures more than 159 billion transactions a year while analysts supply context the models lack; one Brighterion programmer wrote a rule for geographically impossible transactions after a case the models missed. Its Franchise Global Rules Investigation Program gathers evidence on scam merchants and passes it to acquirers.[7][17]
Mastercard describes AI on both sides of fraud. Rigo Van den Broeck, executive vice president of Cybersecurity Solutions, said AI makes card-testing and skimming attacks "repeatable and automated", and a September 2026 Mastercard essay sets the aim of asking whether a consumer is being manipulated into sending money, not only whether a transaction is fraudulent.[22][23]
What is not disclosed: Decision Intelligence Pro results after the February 2024 initial modelling, figures for the foundation model, how many issuers use the Merchant Scam & Risk Indicator beyond the pilot, and fraud-product revenue. The Forrester study page does not state the sample size or who commissioned it.[1][12][5][8][9]
What does the Mastercard profile record on this topic?
The rows below are the entries of the Mastercard AI strategy profile that bear on this page, as of Oct 19, 2026; the profile holds the full record of 16 dated moves, 9 use cases and 23 sources.
| Use case | What the record shows | Status | Source |
|---|---|---|---|
| Decision Intelligence Pro | Generative-AI enhancement to real-time transaction fraud scoring that assesses relationships between entities around a payment; Mastercard reports a 20% average improvement in fraud detection from initial modelling. | Announced | [1] |
| Recorded Future and Mastercard Threat Intelligence | AI-driven threat intelligence from Recorded Future, combined with Mastercard's network visibility in a threat-intelligence service launched in 2025 to detect cyberattacks that lead to payment fraud. | In production | [24][9] |
| Large tabular foundation model | Transaction foundation model trained on billions of anonymized transactions with Nvidia and Databricks; being tested for hybrid cybersecurity and fraud systems and to reduce the thousands of separate models Mastercard maintains. | Pilot | [12] |
| Agentic transaction probability score | A new Agent Pay signal, rolling out for testing in the US, estimates the likelihood that a transaction was initiated by an AI agent so issuers can approve legitimate agent-led purchases with more confidence; built with a Cloudflare data partnership and Skyfire's Know-Your-Agent verification. | Pilot | [13] |
| Date | Move | Detail | Source |
|---|---|---|---|
| Sep 30, 2026 | Agent Pay Trust Framework and new trust and intelligence services | Mastercard published a "Trust for agentic commerce" white paper introducing the Mastercard Agent Pay Trust Framework (five pillars: identity, intent, controls, trusted execution and intelligence) and added new trust and intelligence services to Agent Pay: a probability score estimating whether a transaction was AI-initiated, rolling out for testing in the US, plus a privacy-preserving data partnership with Cloudflare and a partnership with Skyfire to extend Know-Your-Agent verification into Agent Pay. Ann Johnson, EVP of Security Solutions, said "AI agents will make commerce more intuitive, efficient and personal — but only if people and businesses can trust the systems acting on their behalf." | [13][25] |
| Mar 17, 2026 | Large tabular foundation model disclosed | Mastercard describes a generative foundation model trained on billions of anonymized transactions with Nvidia and Databricks; early tests beat standard machine-learning techniques on rare legitimate purchases, and cybersecurity is the first target. | [12] |
| Dec 20, 2024 | Recorded Future acquisition completes | Mastercard completes its purchase of Recorded Future, adding AI-driven threat intelligence to its cybersecurity, identity and fraud-scoring services; the FY2025 10-K puts the cash consideration at $2.7 billion. | [24][9] |
| Feb 1, 2024 | Decision Intelligence Pro announced | Mastercard applies generative AI to its Decision Intelligence fraud-scoring service, which already scores 143 billion transactions a year; DI Pro scans one trillion data points and returns a score in under 50 milliseconds. Mastercard says initial modelling shows fraud detection up 20% on average. | [1] |
Does Mastercard use generative AI to fight fraud?
Yes. Decision Intelligence Pro, announced in February 2024, applies generative AI to one trillion data points at authorisation. In May 2024 Mastercard said generative AI had doubled its detection rate of compromised cards and cut false positives by up to 200%, and in March 2026 it described a transaction foundation model built with Nvidia and Databricks.
How does Mastercard detect scam merchants?
Through Merchant Trust Services and the Merchant Scam & Risk Indicator, announced on May 19, 2026. In a pilot with one issuer the indicator detected about 80% of risky merchants, many up to 90 days early. From July 2026 acquirers must start an investigation within 72 hours when scam activity hits a risk threshold.
Is Mastercard's Consumer Fraud Risk used outside the UK?
Mastercard is taking it abroad through a partnership with Feedzai announced on February 18, 2025, which uses Feedzai's platform in over 90 countries to deploy Consumer Fraud Risk in key markets. In the UK, 15 banks representing 90% of account-to-account payments used it by July 2025.
Are Mastercard's fraud AI results independently verified?
Mostly not. Detection gains for Decision Intelligence Pro and compromised-card detection come from Mastercard's own analysis. Exceptions are a Forrester Total Economic Impact study on Mastercard's site, which reports 12% lower fraud losses for Decision Intelligence & Rules Services users, and UK Payment Systems Regulator scam data Mastercard cites.
Who is liable if an AI agent makes a bad purchase with a Mastercard?
Mastercard has not said. It told Fortune in September 2026 that its Verifiable Intent record of what the cardholder authorized could help resolve a dispute, but did not specify who is responsible if an agent buys outside those instructions. It is working with Visa and Ant International on agent identification.
- Mastercard supercharges consumer protection with gen AI
- Mastercard accelerates card fraud detection with generative AI technology
- Prevent, Protect and Recover: Mastercard strengthens trust in Account-to-Account payments
- Mastercard and Feedzai join forces to protect more consumers and businesses from scams
- How to stop the scammers behind the storefronts (Mastercard Merchant Trust Services)
- Beyond one billion: Inside the journey to a more inclusive and sustainable economy (2025 Impact Report)
- When combating cybercrime, humans need AI — and AI needs humans
- The economic impact of Mastercard Decision Intelligence & Rules Services (undated page, retrieved 2026-10-19)
- Mastercard Incorporated Form 10-K for the fiscal year ended December 31, 2025
- Mastercard leverages its AI capabilities to fight real-time payment scams
- Network International leverages Mastercard's AI-powered Brighterion solution to protect over 60,000 merchants from fraud
- Meet Mastercard's new generative AI model
- Mastercard advances agentic commerce with new trust and intelligence services
- Mastercard Introduces First-Ever Threat Intelligence Solution to Combat Payment Fraud at Scale
- Banks are getting the upper hand on card fraud but social scams getting worse, Mastercard says
- Mastercard Move Transactions Grow 35%, CEO Says Consumer Spending 'Solid'
- Burden of woof: How Mastercard uncovered a $50 million online shopping scam ring
- How payments fraud is growing in scale and sophistication. What companies can do to fight back
- To counter 'friendly' fraud, Mastercard expands technology to new markets
- Payment giants battle the fear factor in agentic commerce
- Card companies are on a 'land grab' to set the standards for AI shopping
- 'The perfect storm': Fighting AI-fueled fraud takes layers of defense
- AI's new role in fighting fraud before it happens
- Mastercard finalizes acquisition of Recorded Future
- What is agentic commerce trust? Inside Mastercard's new framework
When Mastercard moves on AI, you'll read it here first.
when a bank, fintech or regulator moves on AI, it is in the next morning's brief · six sourced stories · 7 am PT · free
plus every regulation, bank and agent page update, the morning after · leave any morning