# Stripe: Fraud and risk AI Source: https://www.bankingnewsai.com/fintech/stripe/fraud-and-risk-ai Last updated: Oct 19, 2026 *Part of the [Stripe AI strategy profile](https://www.bankingnewsai.com/fintech/stripe).* ## How does Stripe use AI to fight fraud, and what results has it published? Stripe's fraud product is Radar, a machine-learning service launched in 2016 and integrated into Stripe's payments platform. It scores each transaction on more than 1,000 characteristics in under 100 milliseconds (April 2025) and draws on the network: Stripe said in 2023 there was a 91% chance a card new to a business had been seen before on Stripe. Stripe's published results run from $4 billion of attempted fraud prevented in 2017, to dispute rates cut 40% in 2021, to more than 20 million card-testing attempts blocked per day at a 2022 peak. In 2024 Radar users' dispute rates fell 17% while industry ecommerce fraud rose 15%, Stripe says. The Payments Foundation Model, announced May 7, 2025, raised detection of card-testing attacks on large users from 59% to 97%. Radar now covers bank debits and, since April 2026, all supported payment methods. The newest target is fraud against AI companies: Stripe says one in six attempted sign-ups at AI services on Stripe is made by a bad actor, and Radar blocked more than 3.3 million risky sign-ups in a month for eight AI businesses. All figures are Stripe's own; no independent measure is published. ([Stripe](https://stripe.com/newsroom/news/stripe-radar-2), [Stripe](https://stripe.com/newsroom/news/radar-for-ach-sepa), [Stripe](https://stripe.com/newsroom/news/enhanced-issuer-network), [Stripe](https://stripe.com/newsroom/news/fraud-trends), [Stripe](https://stripe.com/newsroom/news/card-testing-surge), [Stripe](https://stripe.com/blog/using-ai-optimize-payments-performance-payments-intelligence-suite), [Stripe](https://stripe.com/newsroom/news/sessions-2025), [Stripe](https://stripe.com/us/blog/everything-we-announced-at-sessions-2026), [Stripe](https://stripe.com/newsroom/news/sessions-2026)) ## What results has Stripe published for Radar over time? Stripe's own figures, in the order it published them. Bases differ (attempted fraud, dispute rates, blocked transactions), so they are not one series. | Date | What Stripe reported | Context | Sources | | --- | --- | --- | --- | | April 18, 2018 | Radar prevented $4 billion in attempted fraud in 2017; models trained on hundreds of billions of data points; upgraded models cut fraud by up to an additional 25%; retraining moved to daily | Radar 2.0 and Radar for Fraud Teams | [Stripe](https://stripe.com/newsroom/news/stripe-radar-2) | | May 14, 2021 | Radar blocked 60% more fraud in 2020 than in 2019, and over a billion dollars of fraudulent transactions since launch; an 89% chance Radar had seen a new payment's card before | Bouncer card-scanning acquisition | [Stripe](https://stripe.com/newsroom/news/bouncer) | | July 18, 2022 | Radar reduced dispute rates by 40% in 2021; one May 2022 model update estimated to block $40 million more fraud and recover $70 million more revenue a year | Fraud report on 2019–2021 network data | [Stripe](https://stripe.com/newsroom/news/fraud-trends) | | November 30, 2022 | Blocked more than 20 million card-testing attempts per day at the peak; an additional 400 million fraudulent transactions blocked so far that year; card testing up more than 100x since 2019 | Upgraded models now shipped within an hour | [Stripe](https://stripe.com/newsroom/news/card-testing-surge) | | April 24, 2024 | Dispute rates for Radar users fell 10% in 2023 while online fraud rose 11% | Radar Assistant drafts rules from natural-language prompts | [Stripe](https://stripe.com/newsroom/news/sessions-2024) | | January 15, 2025 | Successful card-testing attacks on Stripe down 80%; stolen-card models blocked an additional 30 million high-risk transactions | Response to verification attacks using stolen-card dumps | [Stripe](https://stripe.com/blog/how-stripe-radar-responded-to-a-new-wave-of-card-testing) | | May 15, 2025 | Dispute rates for Radar users down 17% in 2024 while industry ecommerce fraud rose 15%; DoorDash cut chargeback costs 10% using Radar scores on and off Stripe | Payments Intelligence Suite | [Stripe](https://stripe.com/blog/using-ai-optimize-payments-performance-payments-intelligence-suite) | | December 2, 2025 | More than 24.6 million attempted fraudulent transactions prevented over Black Friday–Cyber Monday (20.9 million worth $917 million in 2024) | Peak-weekend figures | [Stripe](https://stripe.com/newsroom/news/bfcm2025), [Stripe](https://stripe.com/newsroom/news/bfcm2024) | | June 30 and August 19, 2026 | Radar blocked over €2 billion of fraudulent volume for German merchants last year and prevented more than AUD$2.4 billion of fraudulent transactions in Australia in 2025 | Country figures at Stripe Tour Berlin and Sydney | [Stripe](https://stripe.com/newsroom/news/stripe-tour-berlin-2026), [Stripe](https://stripe.com/newsroom/news/tour-sydney-2026) | ## What do Stripe's fraud models use, and how are they built? | Element | What Stripe says | Date | Sources | | --- | --- | --- | --- | | Inputs and speed | More than 1,000 characteristics per transaction; decision in less than 100 milliseconds | April 30, 2025 | [Stripe](https://stripe.com/newsroom/news/radar-for-ach-sepa) | | Score | Each transaction gets a 0–99 risk score; Radar trains on data from millions of businesses, and a card new to a business has a 91% chance of having been seen on Stripe | February 21, 2023 | [Stripe](https://stripe.com/newsroom/news/enhanced-issuer-network) | | Payments Foundation Model | One embedding per payment from a model trained on tens of billions of transactions; classifying sequences of embeddings lifted card-testing detection on large users from 59% to 97% | May 15, 2025 | [Stripe](https://stripe.com/blog/using-ai-optimize-payments-performance-payments-intelligence-suite) | | Training method | Self-supervised learning that discovers its own features; Gaybrick: "generalized models outperform" | May 7, 2025 | [TechCrunch](https://techcrunch.com/2025/05/07/stripe-unveils-ai-foundation-model-for-payments-reveals-deeper-partnership-with-nvidia/) | | Compute | Radar uses NVIDIA's AI platform to improve the speed and accuracy of fraud detection | October 9, 2024 | [Stripe](https://stripe.com/newsroom/news/nvidia-collaboration-with-stripe) | | Issuer feedback | Adaptive rules combine the model's score with the issuer's CVC and postal-code response; businesses that migrate see a 1.3 percentage point rise in payment success | March 27, 2025 | [Stripe](https://stripe.com/blog/using-ai-dynamic-radar-rules) | | Stolen-card detection | Ingests stolen-card data dumps, monitors the internet for card numbers and runs probabilistic stolen-card models | January 15, 2025 | [Stripe](https://stripe.com/blog/how-stripe-radar-responded-to-a-new-wave-of-card-testing) | | Custom models | Custom Radar models trained on a business's own signals plus Stripe's network announced; threshold auto-adjustment during new attacks on the roadmap | April 29, 2026 | [Stripe](https://stripe.com/us/blog/everything-we-announced-at-sessions-2026) | ## How is Radar adapting to fraud against AI businesses? | Item | What the record shows | Date | Sources | | --- | --- | --- | --- | | Trial fraud | Radar expanded to 'friendly fraud' such as free-trial abuse; Stripe says it can stop up to 62% of trial fraud at the source | September 30, 2025 | [Stripe](https://stripe.com/newsroom/news/tour-newyork-2025) | | Free-trial abuse trend | Models detected 6.2x more abusive free trials from November 2025 to February 2026; self-serve AI startups see 10x more attempted abuse than enterprise AI companies | March 24, 2026 | [Stripe](https://stripe.com/blog/how-stripe-radar-helps-prevent-free-trial-abuse) | | Free-trial model results | Predicts abusive trial behavior with 90% accuracy; in the first two months blocked more than 550,000 high-risk trials across four AI businesses, an estimated $4.4 million of compute losses avoided; Cursor is a named user | March 24, 2026 | [Stripe](https://stripe.com/blog/how-stripe-radar-helps-prevent-free-trial-abuse) | | Token theft | One in six attempted sign-ups at AI services on Stripe is by a bad actor; more than 3.3 million risky sign-ups blocked in a month for eight AI businesses | April 29, 2026 | [Stripe](https://stripe.com/newsroom/news/sessions-2026) | | Bot and agent abuse | Preview of bot abuse prevention to distinguish legitimate AI agents from fraudulent actors | April 29, 2026 | [Stripe](https://stripe.com/us/blog/everything-we-announced-at-sessions-2026) | | Stripe Signals | Previews of merchant signals including LLM-powered website reviews, and of Issuing authorization signals extended to non-Stripe issuers, banks and fintechs | April 29, 2026 | [Stripe](https://stripe.com/us/blog/everything-we-announced-at-sessions-2026) | ## How does Stripe handle disputes with AI, and what does it not disclose? Stripe has attached financial guarantees to its models since 2019: Chargeback Protection, built on Radar's machine learning, reimburses disputed amounts and fees for 0.4% per transaction. Smart Disputes (May 2025) uses AI to compile evidence and, Stripe says, helped Vimeo and Squarespace recover 13% more chargebacks; Stripe covers the cost of a countered dispute it loses. ([Stripe](https://stripe.com/newsroom/news/chargeback-protection), [Stripe](https://stripe.com/blog/using-ai-optimize-payments-performance-payments-intelligence-suite)) Radar is sold beyond Stripe's own processing: in April 2024 Stripe said Radar would be available to companies processing with other providers, and in May 2025 it made Radar's risk scores and recommended actions available for payments processed by other PSPs. ([Stripe](https://stripe.com/newsroom/news/sessions-2024), [Stripe](https://stripe.com/blog/using-ai-optimize-payments-performance-payments-intelligence-suite)) What is not disclosed: Radar's current false-positive rate, the Payments Foundation Model's results beyond card testing, results of the custom models announced in 2026, and any independent validation of Stripe's fraud figures. Stripe's 2018 claim of fraud cut "by up to 25 percent" and later percentages are its own measurements on bases it chose. ([Stripe](https://stripe.com/newsroom/news/stripe-radar-2), [Stripe](https://stripe.com/blog/using-ai-optimize-payments-performance-payments-intelligence-suite), [Stripe](https://stripe.com/us/blog/everything-we-announced-at-sessions-2026)) ## What does the Stripe profile record on this topic? | Date | Move | Detail | Sources | | --- | --- | --- | --- | | Apr 29, 2026 | Sessions 2026: 288 launches, agent wallets, token-theft defence | Stripe announced a partnership to let businesses sell inside Google's AI Mode and the Gemini app, Link wallets for agents, streaming payments for AI usage, and Radar coverage of token theft; Radar blocked more than 3.3 million risky sign-ups in a month for eight AI businesses. | [Stripe](https://stripe.com/newsroom/news/sessions-2026), [TechCrunch](https://techcrunch.com/2026/04/30/stripe-link-digital-wallet-ai-agents-shopping/) | | May 7, 2025 | Stripe announces a Payments Foundation Model | At Sessions 2025 Stripe said it had built an AI foundation model for payments, trained on tens of billions of transactions, to be deployed across its payments suite; applied to card testing, it lifted detection of attacks on large businesses by 64%. | [Stripe](https://stripe.com/newsroom/news/sessions-2025) | ## Common questions ### What is Stripe Radar? Radar is Stripe's machine-learning fraud prevention service, launched in 2016 and integrated into Stripe payments. It scores each transaction on more than 1,000 characteristics in under 100 milliseconds, using data from across the Stripe network, and lets businesses add custom rules. Since April 2024 Stripe has offered it to companies that process payments with other providers. ### How much fraud does Stripe Radar stop? Stripe reports figures by period: $4 billion of attempted fraud prevented in 2017; 20.9 million fraudulent transactions worth $917 million blocked over Black Friday–Cyber Monday 2024; over €2 billion blocked for German merchants and more than AUD$2.4 billion in Australia in 2025. These are Stripe's own measurements. ### How did Stripe's foundation model change card-testing detection? Stripe says its earlier models cut successful card testing by 80% over two years, but a sophisticated slice spread across the largest businesses' traffic was hard to catch. Classifying sequences of Payments Foundation Model embeddings raised detection of those attacks on large users from 59% to 97% overnight, Stripe said in May 2025. ### What is token theft at AI companies? Stripe's term for fraud that drains AI compute rather than money: fake accounts that harvest sign-up credits, abuse free trials or run up usage bills they never pay. Stripe said in April 2026 that one in six attempted sign-ups at AI services on Stripe is made by a bad actor, and expanded Radar to score sign-ups and usage in real time. ### Does Stripe Radar work for bank payments? Yes. Stripe extended Radar to ACH and SEPA debits in April 2025, reporting average fraud reductions of 20% on ACH and 42% on SEPA for Radar users, then to Bacs Direct Debit in the UK and, in April 2026, to all supported payment methods, including wallets, buy now, pay later and stablecoins. ## Deeper on Stripe - [How does Stripe use AI across its payment ecosystem?](https://www.bankingnewsai.com/fintech/stripe/payment-ecosystem-ai) - [What is Stripe's role in the digital payments transformation?](https://www.bankingnewsai.com/fintech/stripe/digital-payments-transformation) ## Same question, other companies - [PayPal: fraud and risk ai](https://www.bankingnewsai.com/fintech/paypal/fraud-and-risk-ai) - [Adyen: fraud and risk ai](https://www.bankingnewsai.com/fintech/adyen/fraud-and-risk-ai) - [Visa: fraud and risk ai](https://www.bankingnewsai.com/fintech/visa/fraud-and-risk-ai) - [Mastercard: fraud and risk ai](https://www.bankingnewsai.com/fintech/mastercard/fraud-and-risk-ai) - [Plaid: fraud and risk ai](https://www.bankingnewsai.com/fintech/plaid/fraud-and-risk-ai) - [Amex: How does American Express use AI for fraud detection and credit decisions?](https://www.bankingnewsai.com/banks/american-express/fraud-and-credit-ai) - [BOK Financial: How does BOK Financial use AI against fraud and in credit?](https://www.bankingnewsai.com/banks/bok-financial/fraud-and-credit-ai) - [FNBO: How does FNBO use AI for fraud detection and credit decisions?](https://www.bankingnewsai.com/banks/fnbo/fraud-and-credit-ai) - [MUFG: How does MUFG use AI in credit decisions and against fraud?](https://www.bankingnewsai.com/banks/mufg/fraud-and-credit-ai) - [Pinnacle: How does Pinnacle use AI for fraud detection and credit decisions?](https://www.bankingnewsai.com/banks/pinnacle/fraud-and-credit-ai) - [Sallie Mae: How does Sallie Mae use models and AI in credit and fraud?](https://www.bankingnewsai.com/banks/sallie-mae/fraud-and-credit-ai) - [Western Alliance: How does Western Alliance use AI for fraud detection and credit?](https://www.bankingnewsai.com/banks/western-alliance/fraud-and-credit-ai) --- Canonical page: https://www.bankingnewsai.com/fintech/stripe/fraud-and-risk-ai Part of [BankingNewsAI](https://www.bankingnewsai.com/) — a free daily brief on AI in banking, an AI regulation tracker (42 authorities, 245 documents) and AI-strategy profiles of the 120 largest US banks. Markdown versions of every reference page: append `.md` to the page URL; index at https://www.bankingnewsai.com/llms.txt.