How does PayPal use AI to fight fraud, and what are the results?
PayPal scores risk with machine-learning models that see both sides of its network and reports the outcome as a loss rate. Its 2025 Form 10-K says it uses AI and machine learning in fraud prevention and risk management; transaction losses were about $1.3 billion in 2025, up 20%, which PayPal attributed mainly to fraud incidents affecting PayPal products, while the transaction loss rate held at 0.07% of TPV (0.08% in 2023). In the second quarter of 2026 transaction losses fell 16% to $323 million, a 0.07% rate, on lower fraud losses, risk mitigation strategies and higher recoveries. The named tools are AI-powered scam alerts on Friends and Family payments (July 2025), which use continually learning models that analyze billions of data points; Fraud Protection Advanced, whose machine-learning risk scores draw on PayPal's view of 25 billion transactions a year; and passkeys, which PayPal said in January 2025 cut account-takeover rates by 70% against passwords. Regulators are active: New York DFS fined PayPal $2 million in January 2025 over cybersecurity failures, the FTC sent a civil investigative demand in August 2025 on merchant onboarding, and CFPB demands on Regulation E error resolution are open. PayPal publishes no model accuracy or false-positive figures.[1][2][3][4][5][6]
Data as of Oct 19, 2026 · PayPal profile updated Oct 8, 2026 · 11 sources cited on this page · 52 fintechs profiled on this site · Markdown
Which fraud and risk tools does PayPal run, and what do they do?
| Tool | What the record shows | Date |
|---|---|---|
| Network-wide models | PayPal says its protection rests on its proprietary end-to-end platform and data from both sides of transactions (buyers and sellers, senders and receivers); the 10-K lists fraud prevention and risk management among its AI and machine-learning uses | February 3, 2026[1] |
| Scam alerts for Friends and Family payments | Dynamic alerts at the point of payment, tailored to each transaction's fraud likelihood, getting stricter and adding friction as confidence grows; highly risky payments are still declined automatically; live in all PayPal markets and on Venmo in the U.S. | July 21, 2025[3] |
| Generative AI in fraud prevention | PayPal uses an internally developed equivalent of generative AI to power its fraud prevention system, American Banker reported; Chad Gonzales, VP of consumer and credit fraud risk, said the aim is "running ahead" of new scams | August 7, 2025[7] |
| Fraud Protection Advanced (merchants) | Self-serve tool on PayPal Complete Payments with custom filters, manual review, block and allow lists, and a machine-learning risk score built from card details, buyer information, purchasing patterns and device intelligence | September 14, 2026 (page update)[4] |
| Device signals | Fraud Protection Advanced asks merchants to pass a device ID through PayPal's Fraudnet (browser) or Magnes (app) so it feeds the machine-learning models | September 14, 2026 (page update)[4] |
| Passkeys | PayPal, a FIDO Alliance founding member, reported a 10%+ higher login success rate and 70% lower account-takeover rates on passkey-authenticated transactions than with passwords | January 23, 2025[5] |
| Agentic checkout | Agent ready offers fraud detection and buyer protection on AI surfaces; Perplexity checkouts are screened by PayPal's fraud detection and risk systems | 2025[8][9] |
What do PayPal's loss figures show?
Transaction losses cover customer protection programs, fraud and chargebacks; PayPal reports them against TPV. These are outcomes of the whole risk program, not of AI alone.
| Measure | Value | Period |
|---|---|---|
| Transaction losses | About $1.3 billion in 2025 and $1.1 billion in 2024, up $223 million or 20%, primarily from fraud incidents affecting PayPal products | 2024–2025[1] |
| Transaction loss rate | 0.07% of TPV in 2025 and 2024; 0.08% in 2023 | 2023–2025[1] |
| Transaction loss rate, Q1 2026 | 6 basis points of TPV, improved slightly year over year, which Jamie Miller tied to work on onboarding, fraud prevention and risk management | Q1 2026[10] |
| Transaction losses, Q2 2026 | $323 million, down 16%; loss rate 0.07% against 0.09%, on lower fraud losses, risk mitigation strategies and higher recoveries | Q2 2026[2] |
| Range across reported periods | Transaction loss rate between 0.06% and 0.09% of TPV in the periods in the Q2 2026 10-Q | Periods in the Q2 2026 10-Q[2] |
Which regulator actions touch PayPal's fraud and risk controls?
| Regulator | What | Date |
|---|---|---|
| New York DFS | $2 million penalty under the Cybersecurity Regulation after Form 1099-K data exposed Social Security numbers; DFS found missing controls such as multifactor authentication, CAPTCHA or rate limiting | January 23, 2025[6] |
| FTC | A civil investigative demand in August 2025 on whether merchants' deceptive schemes were facilitated by PayPal's onboarding and due diligence, after a February 2022 CID on commercial customers charging on behalf of other merchants | FY2025 10-K[1] |
| CFPB | Civil investigative demands on Regulation E investigation and error-resolution obligations and on presentment to linked bank accounts, plus an August 2024 CID on PayPal Credit and backup payment options | FY2025 10-K[1] |
| EU AI Act | Cited in the 10-K as a framework that may require operational changes or limit PayPal's use of AI | FY2025 10-K[1] |
Where is PayPal taking risk AI next, and what does it not disclose?
On the Q1 2026 call CFO and COO Jamie Miller said the second wave of PayPal's cost program is aggressive deployment of AI in customer service and operations "equally with respect to risk and the modernization of our risk platform". The 10-Q puts the program at at least $1.5 billion of gross annualized run-rate savings over two to three years.[10][2]
The 10-K and the Q2 2026 10-Q warn that regulators and card networks may adapt error-resolution and chargeback rules to new forms of fraud, which could widen PayPal's exposure and the scope of its protection programs, and that AI-enabled threats are part of the cyber risk it faces.[2][1]
What is not disclosed: model accuracy or false-positive rates, how many scam payments the alerts prevented, results for Fraud Protection Advanced merchants, the share of logins on passkeys, and how much of the 2025 rise in fraud losses came from which products.[3][4][5][1]
What does the PayPal profile record on this topic?
The rows below are the entries of the PayPal AI strategy profile that bear on this page, as of Oct 8, 2026; the profile holds the full record of 15 dated moves, 9 use cases and 20 sources.
| Use case | What the record shows | Status | Source |
|---|---|---|---|
| AI and ML fraud prevention and risk management | PayPal's 10-K says it uses AI and machine-learning technologies in fraud prevention and risk management across its two-sided network. | In production | [1] |
| Agent ready | Lets existing PayPal merchants accept payments on AI surfaces with fraud detection, buyer protection and dispute resolution; announced for early 2026. | Announced | [8] |
| Date | Move | Detail | Source |
|---|---|---|---|
| Feb 3, 2026 | Enrique Lores named CEO; Alex Chriss out | The board appointed Lores, board chair since July 2024 and former HP CEO, effective March 1, 2026, saying progress under Chriss was not in line with its expectations; Lores cited the rapid acceleration of AI reshaping commerce. The 2025 10-K, filed the same day, describes AI use in fraud, risk, product optimization and customer service. | [11][1] |
| Oct 28, 2025 | Agentic commerce services launched | PayPal introduced store sync, to make merchant catalogs discoverable in AI channels such as Perplexity, and agent ready, to let existing merchants accept payments on AI surfaces with fraud detection and buyer protection (planned for early 2026). | [8] |
| Jan 23, 2025 | New York DFS fines PayPal $2 million over cybersecurity | DFS found PayPal failed to use qualified personnel for key cybersecurity functions and lacked controls such as multifactor authentication, CAPTCHA or rate limiting, after cybercriminals used compromised credentials to reach Form 1099-Ks containing Social Security numbers; PayPal has since remediated the issues, DFS said. | [6] |
Does PayPal use AI to detect fraud?
Yes. PayPal's 2025 Form 10-K says it uses AI and machine-learning technologies in fraud prevention and risk management. Named tools include AI-powered scam alerts on Friends and Family payments, launched in July 2025, and Fraud Protection Advanced, which gives merchants a machine-learning risk score drawing on 25 billion transactions a year.
How much does PayPal lose to fraud?
PayPal does not report fraud losses alone. Transaction losses, which include protection programs, fraud and chargebacks, were about $1.3 billion in 2025, up 20% mainly on fraud incidents, a rate of 0.07% of TPV. In Q2 2026 they fell 16% to $323 million, also 0.07% of TPV.
How do PayPal's scam alerts work?
When the system detects a likely scam on a Friends and Family payment, an alert appears before money is sent, tailored to that payment's fraud likelihood. Alerts get progressively stricter as confidence grows, and highly risky payments are still declined automatically. PayPal says the AI models analyze billions of data points and update as patterns change.
Do passkeys reduce account takeover at PayPal?
PayPal said in January 2025 that account-takeover rates for transactions authenticated with a passkey were 70% lower than with passwords, and login success rose by more than 10%. PayPal, a founding member of the FIDO Alliance, said it planned to accelerate passkey availability in 2025.
Which regulators have acted on PayPal's risk controls?
New York DFS fined PayPal $2 million in January 2025 under its Cybersecurity Regulation. The 10-K discloses FTC civil investigative demands from 2022 and August 2025 on merchant onboarding and due diligence, and CFPB demands on Regulation E error resolution and PayPal Credit. PayPal says it is cooperating.
- PayPal Holdings, Inc. Annual Report (Form 10-K) for fiscal year 2025
- PayPal Holdings, Inc. Form 10-Q for the quarter ended June 30, 2026
- Introducing AI-Powered Scam Alerts for Friends and Family Payments
- Fraud Protection Advanced (developer documentation, last updated September 14, 2026)
- Solving the Convenience and Security Equation
- Superintendent Adrienne A. Harris Secures $2 Million Cybersecurity Settlement with PayPal, Inc.
- PayPal and Visa embrace AI for payment crime fighting
- PayPal Launches Agentic Commerce Services to Power AI-Driven Shopping
- PayPal and Perplexity Launch Instant Buy Ahead of Black Friday
- PayPal (PYPL) Q1 2026 earnings call transcript (The Motley Fool)
- PayPal Appoints Enrique Lores as Chief Executive Officer and David W. Dorman as Independent Board Chair
When PayPal 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