Fraud Classification
- First party: the person is lying. Own identity, own details, false claim. Confusion and valid complaints aren't lies; this site keeps them out
- Second party (this site's convention): they handed it over. Mule, family member, or someone talked into it
- Third party: nobody gave them anything. It was taken
- Synthetic: a fabricated person or entity. Real people's identifiers can be inside it, and those people can be victims
- Your response differs: eligible 3DS payments for third party, evidence for friendly fraud, payout controls for second party
- The industry doesn't agree on any of this. The main competing definitions are below, each beside its source
A dispute lands, or a vendor's report uses a fraud label you don't recognize. Here's how to tell which kind of fraud you're looking at, and what each one needs from you.
Quick Classification
| What You See | Likely Type | Your Response |
|---|---|---|
| Customer disputes a purchase they really made, with no valid reason | Friendly Fraud | Collect evidence. On Visa, CE 3.0 may let earlier purchases answer it |
| Card used by someone who was never given access | Third-Party Fraud | Enable 3DS; liability can shift on eligible payments |
| Same device, many accounts | Fraud Ring | Device fingerprinting, block infrastructure |
| Burst of small transactions | Card Testing | Velocity rules, CAPTCHA |
| Good customer acting strange | Account Takeover | Add MFA (two different kinds of factor, not two steps of the same kind), and flag logins that don't match the customer's usual behavior |
| Excessive returns/refunds | Refund Fraud | Policy enforcement, tracking |
| Promo/coupon abuse | Promo Abuse | Device linking, limits |
| Fake account signups | Account Fraud | Email/phone verification |
| Mismatched identity info | Fake Identity | Identity verification |
The Four Types, Defined
| Type | The test | In one line |
|---|---|---|
| First party | Is the person lying? | They're using their own identity, not hiding who they are, and lying about something |
| Second party | Did they hand it over? | The real owner gave their details, card or account to someone else |
| Third party | Were they authorized? | Somebody who was never given access is using it |
| Synthetic | Is this a real person? | A fabricated person or entity, assembled from identifying information, real, fake, or both |
First party: the person is lying
They aren't hiding who they are. Their name is their name, their address is their address. The lie is about something else.
Two shapes, and they're the same act:
- Lying about facts. Inflating income on an application, misstating what a business does
- Lying about an event. Saying they didn't make a purchase they made
That second one is the merchant-facing version, usually called friendly fraud. A customer who forgot a purchase, or has a fair complaint, hasn't lied. This site keeps confusion and valid disputes out of first party; some vendors group them in, which is one reason their numbers differ.
A customer who borrows and can't pay hasn't lied. That's credit risk, and it's a pricing problem rather than a fraud problem. A customer who lied to get the money committed fraud. The lie is the line. Experian's definition below, for one, requires an intent not to repay when the promise is made. Owing money you later can't pay doesn't meet it.
Second party: they handed it over
Someone gave their identity, card, account or device to another person. The details are genuine because they really are genuine. The owner participated.
This handover test is this site's convention. It classifies the case; it doesn't settle who authorized a particular payment, or who bears the loss. Record those separately: who initiated the payment, what authority they had for that use, and whether deception or coercion was involved. Credentials phished and then used by someone else weren't an authorized payment just because they changed hands.
Why they did it varies enormously, and the reason doesn't change the classification:
| Situation | Still second party? |
|---|---|
| Paid to let their account move money (a money mule) | Yes |
| Lent a card to a parent, sibling or partner | Yes |
| Talked into it on the phone, or by a fake job ad | Yes |
| Pressured or controlled by someone else | Yes |
| An employee abusing internal access | No. That's insider fraud, and it goes under first party |
Second party covers people who were paid and people who were tricked. Same classification, completely different culpability. Say which one you're looking at, separately.
Somebody deceived by a phone call still handed their details over. That's second party. They're also a victim. Both are true.
Third party: nobody gave them anything
The person using the card or identity was never given access to it. It was taken, not handed over. Stolen card data, a breached account, a hijacked login.
Under this site's convention, that's the difference from second party. Did the owner hand it over, or did somebody take it?
Synthetic: a fabricated person
A fabricated person or entity, assembled from identifying information. The composite persona doesn't exist, but real people's identifiers can be inside it, and those people can be victims. Work out whose details were used instead of assuming nobody is affected. (Federal Reserve, 2019 review, pages 5 and 17.)
The Fed's 2019 paper names three ways they're built:
| Term | What it is |
|---|---|
| Fabrication | Built from wholly fictitious identifying data |
| Manipulation | Real identifying data altered |
| Compilation | Real and fictitious identifying data combined into a new identity |
SentiLink's 2020 proposal cuts it differently, into first-party and third-party synthetics. That's SentiLink's cut, not the Fed's. (SentiLink, Defining Synthetic Fraud, 2020, pages 5 to 6.)
Worked example: "my family member used my card"
This is an ambiguous case merchants see, and it lands in a different class depending on one fact. Filing all of it as friendly fraud hides that fact, and it changes what you should do.
| What actually happened | Class | What you do |
|---|---|---|
| Cardholder gave the card to a relative, then disputed the charge | First party, if the denial is knowing | Answer with your evidence, including who used the card if you can show it |
| Cardholder gave the card to a relative, and genuinely forgot | Confusion, not fraud on these facts | Refund or answer on cost, and fix your descriptor. Don't log it as fraud |
| Cardholder knowingly let a relative use it to move money | Second party | Treat as mule risk, not a dispute |
| Relative took the card without asking | Third party | A real unauthorized use. The cardholder is a victim |
Amex names this on its own merchant material, describing charges made by "family or staff." Some of what merchants log as fraud is a second cardholder in the same house or business, and some of that isn't fraud at all.
The question to ask, and it's one question: did the cardholder hand the card over?
Does the Rest of the Industry Agree?
Closer on first, third and synthetic. Not on second. Here's what the sources say.
Visa's rules use first-party fraud, in issuer customer validation (24 October 2026 edition, section 10.4.7.3). When someone says "the networks define it this way," ask which document.
Where the numbering comes from
One way to see the confusion, as an analogy rather than a history: a normal transaction has two parties, you and your customer.
| Party | Who that is |
|---|---|
| First party | your customer |
| Second party | you, the merchant |
| Third party | an outsider who was never in the transaction |
Read that way, "second-party fraud" would mean fraud by the merchant. That's not how the sources below use it. Whether the numbering started here isn't documented in them; treat the table as an illustration, and the disagreements below as the evidence.
First party: two camps, one act
Lenders describe the application. The Experian and DataVisor wording below is as reproduced in SentiLink's 2020 paper, page 11.
| Source | What they say |
|---|---|
| Experian | "an individual who makes a promise of future repayment in exchange for goods / services without the intent to repay" |
| TransUnion | "misrepresenting one's personal information in order to receive a product or service to which they have no entitlement." Also calls it "credit muling" |
| Equifax | "an individual intentionally misrepresents their identity and/or gives false information for financial or material gain" |
| Socure | "the use of one's own identity to open an account and/or commit a dishonest act for personal or financial gain" |
| DataVisor | "the fraudster is not misrepresenting who they are, but rather, they're being deceptive about their information, and their intentions" |
Payments people describe the dispute.
| Source | What they say |
|---|---|
| Ethoca (Mastercard) | uses first-party fraud and friendly fraud as overlapping names for a customer disputing a genuine purchase |
| Mastercard | "When cardholders dispute a genuine transaction, that's first-party misuse or 'friendly' fraud" |
| Signifyd | defines it around a real identity used to defraud an online merchant, and separates deliberate deception from mistaken disputes |
Those look like two definitions. They're the same test applied at two moments: own identity, false statement. Lying about your income and lying about a purchase are both lying. That's why we test for the lie instead of picking a camp.
It matters when you read a number. "First-party fraud is up 40%" tells you nothing on its own. Ask what it counts: loan applications, disputed charges, or both. Visa's commissioned research concedes it: "This abuse of the dispute system has many names, such as first-party fraud, friendly fraud, and dispute misuse. Regardless of what it is called..."
SentiLink has said it plainly in print. First: "First Party Fraud has differing definitions depending upon who or what company you ask." Then: "The definition of First Party Fraud is confusing and can be interpreted both as (1) credit risk or (2) fraud risk."
Second party: five different definitions
Every source below describes something real. They disagree about how willing the person had to be.
| How willing was the account holder? | Example | Who calls this second party |
|---|---|---|
| Paid volunteer. Knew, took a cut | classic money mule | TransUnion, SEON, Persona, FraudNet |
| Helping someone. Knew, wasn't paid | lending a card to a relative | Linkurious, Persona |
| Tricked. Handed details over believing a lie | phishing, fake job ad | This site (see below) |
| Coerced. Under someone else's control | trafficking, an abusive relationship | Alloy |
| Not the account holder at all | an employee abusing access | AU10TIX, NordVPN |
Some of the wording, so you can see the gap:
| Source | What they say |
|---|---|
| Persona | describes paid sharing and helping family, with both people as active participants |
| Alloy | describes one person controlling another to commit fraud |
| AU10TIX | describes collusion between an insider and an external party |
| NordVPN | describes a perpetrator with an existing relationship to the target, or an insider |
Two of those point at opposite people. For TransUnion and Persona, the second party is the customer who handed their details over. For AU10TIX, it's the employee.
And "knowingly" is doing a lot of work. Definitions built on willing complicity drop the person tricked by a fake job ad and the person under someone else's control. They handed over the same details, and the money moves the same way. We keep them in, and treat classification and culpability as separate questions.
We take the first four rows and reject the fifth. Handing something over is one act, whatever the reason. It's also something you can observe. An employee abusing access never handed over anything of their own. Insider fraud already has a mature home: the ACFE (Association of Certified Fraud Examiners) covers it as occupational fraud.
Third party: the closest to agreement
| Source | What they say |
|---|---|
| Experian | describes use without the owner's consent or knowledge |
| TransUnion | "the person whose information is being used is unaware and has not given consent" |
| Ethoca (Mastercard) | describes an unauthorized person using someone's card information |
| AU10TIX | describes an external actor, with phishing among its examples |
Unauthorized is the word that keeps recurring, and it's the test we use. Even here conventions differ: Alloy groups some authorized-push-payment scams under third party. This site's handover line is a choice, not a consensus.
Synthetic: defined, and filed by the Fed's model
The Federal Reserve put 12 fraud experts in a room to settle it. They landed on this: "the use of a combination of personally identifiable information (PII) to fabricate a person or entity in order to commit a dishonest act for personal or financial gain." (Recommended definition, 2021, page 4.)
SentiLink: "fraudsters combine fictitious and/or real information to create new identities with the intent to defraud."
The definition itself doesn't put synthetic in a party bucket. The Fed's separate FraudClassifier model does classify payments, and it files Synthetic ID under "Authorized Party Acted Fraudulently," with a fabricated-account example.
Vendors file it differently too; Linkurious is among those folding it toward first party. This site keeps synthetic as its own category. That's our convention, not the Fed's, and it doesn't decide who bears a loss.
Some vendors aren't sorting by actor at all
One reason two lists don't line up:
Everything above sorts by who did it. Some vendors sort by when it happened instead: at the application, at the transaction, or on the account afterwards.
An application-fraud bucket, "lying on the form," can hold three different actors at once:
| Inside one application bucket | Where we put it |
|---|---|
| Stolen identity | Third party |
| Synthetic identity | Its own class |
| First-party fraud | First party |
So one of their classes contains three of ours. Neither list is wrong. They answer different questions, and a term that's a subtype in one is a top-level class in the other.
What to do with that. When a vendor's taxonomy doesn't match yours, check the axis before you argue about definitions. Ask: is this sorted by who, or by when? A lot of what looks like disagreement is one list sorted by actor and another sorted by stage, set side by side as though they measured the same thing.
Ask one question: does that figure count loan applications, disputed charges, or both?
LexisNexis reported first-party fraud as 36% of reported fraud in its Digital Identity Network's 2024 observations, across more than 104 billion platform transactions of many kinds: account creation, logins, payments and more. That's a different population from a chargeback-abuse figure. Compare definition, coverage, denominator and period before comparing percentages.
By Actor
First-Party Fraud (Your Customer)
The customer is real and uses their own identity, but abuses your policies.
| Subtype | Description | Defense |
|---|---|---|
| Friendly Fraud | Disputes a purchase they really made, with no valid reason | Evidence collection, CE 3.0 |
| Refund Fraud | Exploits return policies | Policy enforcement |
| Promo Abuse | Games promotions/discounts | Device linking, limits |
| Insider fraud | An employee abuses their access | Role-based access, reviews |
Chargeback outcome: Can be won with proper evidence
Second-Party Fraud (Your Customer, Working For Someone Else)
The real owner handed over their details, card or account. The identity is genuine because it really is genuine. The money goes to somebody else. Merchants can meet this as a marketplace or payout problem rather than at checkout.
| Subtype | Description | Defense |
|---|---|---|
| Money mule | Account exists to receive and forward funds | Payout pattern review, no-trade sellers |
| Family and friends | A relative uses the card with permission | Ask before treating it as fraud |
| Talked into it | Fake job ad, phone scam, romance scam | Treat the holder as a victim |
| Account farming | Aged accounts sold on for their history | Dormant-then-active review |
Chargeback outcome: can show up as account closure and payout risk rather than a card dispute
Someone talked into handing over their details is second party, because they handed them over. They're also a victim. Classification and culpability are separate questions, and you should answer both.
Third-Party Fraud (Taken, Not Given)
Somebody who was never given access is using the card or identity. It was taken, not handed over. If the owner handed it over, however they were persuaded, that's second party instead.
| Subtype | Description | Defense |
|---|---|---|
| Stolen Card | Uses compromised card | 3D Secure, AVS/CVV |
| Card Testing | Validates stolen cards | Velocity rules, CAPTCHA |
| Account Takeover | Hijacks customer account | Add MFA (two different kinds of factor, not two steps of the same kind), and flag logins that don't match the customer's usual behavior |
Chargeback outcome: Hard to win without liability shift on an eligible 3DS payment
Fake Identity Fraud
Fraudster creates fabricated or mixed identity information.
| Subtype | Description | Defense |
|---|---|---|
| Fake Identity | Fabricated persona | Identity verification |
| Account Fraud | Fake account signups | Email/phone verification |
Chargeback outcome: depends on the network condition. An identity mismatch alone doesn't win it; prove the facts that condition requires
Organized Fraud
This one is a different question. The classes above ask who the actor was. This asks how many. Several accounts are a pattern to investigate, not proof of a ring: a ring is an organized group, and one person can run many accounts.
| Subtype | Description | Defense |
|---|---|---|
| Fraud Rings | Multi-account attacks | Device fingerprinting, consortium data |
| Triangulation | Resale scheme using a fake storefront | Shipping address analysis |
Chargeback outcome: Document network evidence for representment
Classification Decision Tree
Treat this as a set of investigative hypotheses. Actor, identity construction, compromise mechanism and coordination are separate questions. They can overlap, and "unknown" is an allowed answer. A synthetic account can be taken over; one actor can run several accounts.
Response by Classification
| Fraud Type | Immediate Action | Prevention | Chargeback Strategy |
|---|---|---|---|
| Friendly Fraud | Collect delivery proof | Clear descriptors, communication | The evidence your network condition allows; CE 3.0 where it applies |
| Second-Party | Review payouts, close if confirmed | Payout change controls | Answer inside the actual dispute, if one is filed |
| Third-Party | Cancel if caught. If it's already disputed, answer inside that dispute instead of refunding separately | Enable 3DS | Eligible 3DS liability shift, where its conditions are met |
| Card Testing | Block IP/device | Velocity limits, CAPTCHA | Successful test payments can be disputed; answer those cases |
| ATO | Lock account, notify customer | MFA, device recognition | Compromise alone isn't a defense; use the defense your network condition permits |
| Refund Fraud | Flag account | Enforce policies | Document abuse pattern |
| Promo Abuse | Revoke benefits | Device linking | Handled on the account, unless a disputed charge is involved |
| Fraud Ring | Block infrastructure | Device fingerprinting | Document the pattern the network condition asks for |
Financial Trouble Versus Deliberate Abuse
A share of what gets counted as first-party fraud isn't fraud. It's people who can't pay.
- LexisNexis runs a whole category called "distressed consumers," separate from fraud
- Visa's commissioned research found consumer motivations split roughly evenly between wanting to avoid payment and facing economic hardship (2,276 surveyed consumers, Q2 2024, in the markets covered)
- LexisNexis again: first-party fraud is "exacerbated by periods of inflation and the rising cost of living"
One more finding, before you build rules around age. The behavior is "not limited to younger demographics."
That matters for policy, but how a customer responds to a payment plan doesn't prove what they intended. Weigh authority, whether the claim is valid, account history, and what each remedy costs. Keep hardship, mistakes and deliberate deception as separate hypotheses until the facts separate them.
Related Topics
- Reference - The whole reference section
- Fraud Types Overview - Detailed guides for each type
- First-Party Fraud - Customer abuse in detail
- Third-Party Fraud - Used without consent
- Friendly Fraud - Customer disputes
- Fake Identity Fraud - Fabricated personas
- 3D Secure - Liability shift for third-party fraud
- Device Fingerprinting - Linking fraud cases
- Compelling Evidence - Fighting chargebacks