Identity theft announces itself through a victim who complains, while synthetic identity fraud has no victim to complain, so a fraud operation built around customer disputes will catch one and keep lending to the other. Both start with personal data that does not belong to the applicant, but they differ in who is harmed, how long they run, which signals expose them, how the loss is booked and which team should respond. This guide compares the two side by side: threat model, lifecycle, detection signals at onboarding and on the book, the effect of generative AI documents, legal obligations that apply to one but not the other, ownership, and a decision checklist. It is for fraud strategy, identity proofing, credit risk and financial crime teams. Detail on each fraud type sits in the Synthetic Identity Fraud Guide and the Identity Theft Recovery Guide; this page covers the decisions where the two meet.
Key takeaways
- The victim is the dividing line. Identity theft uses a real person's identity and harms that person, so it is usually reported quickly. A synthetic identity is a fabricated person, often built on a child's or elderly person's identifier, so the fraud is often never reported at all.
- Different signals, different moments. Identity theft shows up as a mismatch between the applicant and the real owner's history, and later as disputes. Synthetics show up as an identity that is internally consistent but too new, too thin or linked to other identities.
- The loss lands in different ledgers. Identity theft losses are booked as fraud because a victim says so. Synthetic losses often look like credit charge-offs, which hides the network behind them.
- The law follows the victim. US identity theft victims have rights to blocks, records and Red Flags programme responses; a synthetic has no consumer to exercise them, so obligations shift to suspicious activity reporting and portfolio clean-up.
- Generative AI blurs the onboarding signals for both. Fabricated or altered document images and injected video now support both stolen and synthetic identities, so document and liveness checks need injection controls, not only visual inspection.
The short answer
If the identity belongs to one real person and someone else is using it, it is identity theft. If the identity does not correspond to any single real person, because it fabricates, manipulates or combines personal data, it is synthetic identity fraud. The Federal Reserve's industry definition is "the use of a combination of personally identifiable information (PII) to fabricate a person or entity" to commit a dishonest act. FinCEN classes synthetic identities under impersonation, alongside other false-identity schemes, while account takeover belongs to compromise.
| Dimension | Identity theft (true-name fraud) | Synthetic identity fraud |
|---|---|---|
| Identity used | One real person's full identity | Fabricated, manipulated or compiled identity that matches no single real person |
| Direct victim | The person whose identity was used, plus the lender | The lender or programme; sometimes a child or elderly person whose identifier was used, who may not find out for years |
| How it is usually found | Victim notices accounts, bills, credit file entries or collection calls and complains | Default with no contact, link analysis, or a bust-out across linked accounts |
| Typical duration | Weeks to months before discovery | Months to years of cultivation before bust-out |
| Credit file | Real, often long history that the fraud piggybacks on | Created by the fraudster's own applications; thin, young or boosted by authorised user tradelines |
| Where loss is booked | Fraud loss, confirmed by the victim | Often a credit loss; the Federal Reserve notes institutions differ in booking it as credit or fraud |
| Customer contact | The real person wants the account closed and the record cleaned | Nobody to contact; the "customer" disappears |
| Natural response owner | Fraud operations with customer care and disputes | Fraud strategy with credit risk and financial crime |
Identity theft: threat model and lifecycle
True-name fraud uses data from breaches, phishing, mail theft or social engineering to apply for credit, open deposit accounts, file tax returns or take over existing relationships in the victim's name. The FTC received more than 1.1 million identity theft reports through IdentityTheft.gov in 2024, a reminder that victims do report at scale.
- Acquisition: the fraudster holds the victim's name, date of birth, identifier and often a document image or answers to knowledge questions.
- Application: the application passes data validation because the data is genuine. The weak points are contact details (new phone, email or address that do not match the victim's history), device and location, and whether the applicant can prove they are the rightful owner of the evidence.
- Use: rapid draw-down, card shipping to a new address, or immediate transfers out. The Federal Reserve notes traditional identity fraud "is typically detected and reported relatively quickly" because the victim is directly affected.
- Discovery and dispute: the victim contacts the lender, files a report and asks for blocks. This triggers the legal duties covered below.
The defining strength for defenders is that a real person exists whose history can be compared with the applicant, and who will tell you when something is wrong. For prevention at onboarding see the Identity Fraud Prevention Guide.
Synthetic identity fraud: threat model and credit bust-out
A synthetic identity is assembled from fabricated data, slightly altered real data or a mix of both, then cultivated. The Federal Reserve's 2019 white paper describes the pattern: repeated applications across institutions until one creates a credit file, a thin file built up with small credit lines and on-time payments, piggybacking as an authorised user on established accounts, then the bust-out, which the paper defines as "maxing out the credit line and vanishing". Fraudsters may also claim identity theft to remove charges, or pay balances down with bad cheques before maxing out again, then start over with a new synthetic.
- Victimless on the surface: where a real identifier is used it typically belongs to someone unlikely to check a credit file, which is why the fraud is often unreported. See the Child Identity Theft Guide for the overlap.
- Model blind spot: the white paper cites an industry estimate that 85 to 95 per cent of applicants flagged as potential synthetics were not flagged as high risk by traditional fraud models, because those models expect a real person behind the application.
- Uses beyond credit: FinCEN's deepfake alert describes synthetic identities used to open accounts that receive and launder proceeds of other fraud, so a synthetic may act as a mule account rather than a borrower. See the Money Mule Detection Guide.
- Scale indicators: in FinCEN's analysis of 2021 Bank Secrecy Act data, synthetic identity was a small labelled typology (about 3,000 reports, roughly $182 million), which says as much about under-labelling as about volume.
Detection signals compared
The two fraud types fail different tests. Identity theft fails "is this applicant the rightful owner of this identity?"; a synthetic fails "does this identity belong to one real person with a history that fits?".
| Signal family | Points to identity theft | Points to a synthetic identity |
|---|---|---|
| Core data validation | Name, date of birth and identifier validate against authoritative sources | Identifier fails or only partly matches; one identifier seen with several names or dates of birth; randomised or unissued identifiers alongside an older claimed age |
| Contact details | New phone, email or address that conflicts with the real person's long history; recent SIM swap | Contact details with no history at all; drop addresses such as mail boxes or vacant properties |
| Credit file shape | Long, established file; applicant behaviour out of character for it | Thin or young file; history dominated by authorised user tradelines; depth that does not fit the claimed age |
| Linkage | Device or IP shared with other takeover or theft cases | Phone, address, device or employer shared across identities that claim to be different people |
| Evidence and biometrics | Stolen genuine document; face does not match the document holder | Fabricated or altered document; face matches the document because both are fake |
| Post-onboarding behaviour | Rapid draw-down, change of address, new payees soon after opening | Perfect payment record, rising limits, then simultaneous maxing out across products and lenders |
| Customer contact | Victim calls, disputes or files an identity theft report | No contact; later false identity theft claims on charged-off accounts |
The practical consequence is that the strongest identity theft controls are ownership checks (document to face matching, possession of the real person's contact channels, authoritative data checks), while the strongest synthetic controls are existence and linkage checks across the whole portfolio. NIST SP 800-63A-4 reflects this: it says IAL1 is designed to limit highly scalable attacks and synthetic identities, while IAL2 adds checks that the applicant is the rightful owner of the evidence. It also states that knowing an SSN "is not sufficient to act as evidence of identity". See the NIST SP 800-63A-4 Identity Proofing (IAL) Guide.
AI-generated documents and injected media
Generative AI changes the evidence step for both fraud types. FinCEN alert FIN-2024-Alert004 (13 November 2024) describes criminals altering or generating photos for driving licences and passport cards and combining them with stolen or fabricated personal data. For identity theft, the fake image lets a fraudster pair a real person's data with their own face; for synthetics, it supplies a document for a person who does not exist.
- Red flags from FinCEN: a photo internally inconsistent or mismatched with other identifying information, multiple documents inconsistent with each other, a third-party webcam plugin during live verification, attempts to change communication method during a check, refusing MFA, a reverse image match to galleries of AI-generated faces, and geographic or device data inconsistent with the documents. FinCEN asks for the key term FIN-2024-DEEPFAKEFRAUD in suspicious activity reports.
- Controls from NIST: SP 800-63A-4 requires credential service providers to apply technical controls that increase confidence media comes from a genuine sensor (for example detecting virtual cameras and emulators), to analyse submitted media for manipulation, to use live document capture with presence checks, and, for remote biometric capture, presentation attack detection tested to ISO/IEC 30107-3:2023.
- What changes in the comparison: a face that matches its document used to argue against identity theft. With generated documents, a match proves only that both images came from the same source, so linkage and existence checks matter more for both fraud types.
For capture-side defences see the Biometric Injection Attacks guide and Document Verification vs Liveness Detection.
Legal obligations that differ
In the US, consumer protection law is built around the identity theft victim. Several duties apply only when a real person comes forward.
| Obligation | Identity theft | Synthetic identity |
|---|---|---|
| FTC Red Flags Rule, 16 CFR 681.1 | Written programme to identify, detect, respond to and periodically update Red Flags for covered accounts; board or committee approves the initial programme | Same programme applies at onboarding; suspicious documents and personal information categories in Appendix A catch some synthetics |
| FCRA 605B block, 15 U.S.C. 1681c-2 | Consumer reporting agency blocks reported items within 4 business days of proof of identity, an identity theft report and a statement, and notifies the furnisher | No real consumer to request a block; a false identity theft claim on a synthetic account can be declined or rescinded on grounds of misrepresentation |
| FCRA 609(e), 15 U.S.C. 1681g(e) | Business must give the victim application and transaction records within 30 days of a qualifying written request, at no charge | Not triggered unless a real person whose identifier was used comes forward |
| Suspicious activity reporting (FinCEN) | Applies where thresholds are met | Main reporting route; describe the synthetic typology in the narrative and add the deepfake key term where media was manipulated |
This guide summarises US obligations as published on the official sources in November 2026 and is not legal advice. For the programme detail see the FTC Red Flags Rule guide.
Who owns the response
Ownership should follow the decision that has to be made, not the label on the case.
| Activity | Identity theft | Synthetic identity |
|---|---|---|
| First responder | Fraud operations, often triggered by customer care or disputes | Fraud strategy or investigations, triggered by models, link analysis or collections |
| Customer handling | Victim support, account closure, record correction, 609(e) records | None; close linked accounts, stop limit increases, restrict authorised user additions |
| Credit reporting | Correct or delete furnished data; honour 605B blocks | Stop furnishing positive history for the synthetic; reclassify from credit to fraud loss |
| Network follow-up | Check device, contact and payee links to other victims | Search the book for every account sharing identifiers, contacts, devices or employers |
| Reporting | Law enforcement referral where useful; SAR where required | SAR; consortium data sharing; internal loss classification |
| Model feedback | Label as third-party fraud with the identity step exploited | Label as synthetic with construction type, so credit and fraud models learn from it |
The joint duty is loss classification. If collections can write off a synthetic as a credit loss without fraud review, the network survives. Agree a rule: any charged-off account with no customer contact, a thin or authorised user dominated file, or links to other charge-offs goes to fraud review before the loss is booked. Case handling patterns are in the Fraud Operations and Case Management Guide.
Worked scenarios
- Scenario 1, the complaint: a 54-year-old customer calls about a card she never opened. Her name, date of birth and identifier validated; the application used a new email, a mobile number first seen that week and a delivery address 300 miles away. This is identity theft. Fraud operations closes the account, corrects furnished data, prepares 609(e) records on request and searches for other applications using the same phone and device.
- Scenario 2, the quiet charge-off: a card approved two years ago on a thin file paid perfectly, had three limit increases, then maxed out in one week along with two other cards at the same institution. No one answers collections calls. The identifier was issued after the claimed date of birth would suggest, and the phone number is shared with five other customers. This is a synthetic. Fraud strategy reclassifies the loss, stops furnishing positive history, and reviews the five linked accounts before they bust out.
- Scenario 3, the false claim: after a charge-off, an identity theft dispute arrives for an account flagged as synthetic. The dispute is routed to fraud investigation rather than processed as a routine claim, because the Federal Reserve describes false identity theft claims as a way to double the payout.
Decision checklist for classifying a case
- Does the core identity (name, date of birth, identifier) validate as one real person in authoritative sources? If no, treat as synthetic until shown otherwise.
- Has a real person come forward to say they did not open the account? If yes, treat as identity theft and start the victim process.
- Does the identity's history fit its claimed age, and is it dominated by authorised user tradelines? Thin, young or piggybacked files point to synthetics.
- Do contact details, devices or employers link this identity to others that claim to be different people? Links point to a synthetic ring or a theft crew; check both.
- Was the document or selfie flagged for manipulation or injection? Record it for both fraud types and add the deepfake key term to any SAR.
- Was there customer contact before loss? None, plus a bust-out pattern, points to synthetic and must go to fraud review before a credit loss is booked.
Measuring both
- Synthetic share of charge-offs: proportion of credit charge-offs reclassified as synthetic fraud after review; a rising number at first usually means better classification, not more fraud.
- Linked-account catch rate: linked synthetic accounts closed or restricted before they bust out, as a share of all linked accounts found.
- Victim-reported identity theft rate: confirmed identity theft per 10,000 new accounts, by channel, and the share caught before the victim reported it.
- Time to resolution for victims: days from report to account closure, record correction and 609(e) records sent.
- Onboarding false positive rate: good applicants declined or stepped up per confirmed fraud prevented, measured the same way for both fraud types.
Common failure patterns
- Relying on victim complaints: a programme that only investigates disputes will never see synthetics, which by design generate none.
- Booking synthetic losses as credit risk: the network stays open and keeps cultivating new accounts.
- Treating a matching face as proof: generated documents and injected video can match each other perfectly.
- Processing false identity theft claims as routine: disputes on accounts already flagged as synthetic should go to investigation, not straight to write-off.
- Splitting data across teams: identity theft cases in customer care systems and synthetic cases in credit risk systems mean the shared phones and devices are never linked.
- Using knowledge-based questions as verification: SP 800-63A-4 says knowledge-based verification shall not be used for identity verification, and both fraud types can answer them.
Practitioner checklist
- Define identity theft and synthetic identity in your loss taxonomy, using the Federal Reserve construction types for synthetics.
- Route every no-contact charge-off with a thin or authorised user dominated file to fraud review before the loss is booked.
- Run link analysis on phones, addresses, devices and employers across the whole book, not only at onboarding.
- Validate core identity data against authoritative sources, including a death records check.
- Apply injection, genuine sensor and presentation attack controls to document and selfie capture.
- Keep a documented process for 605B block notices and 609(e) record requests with service levels.
- Route identity theft claims on suspected synthetic accounts to investigation.
- Describe synthetic identity typologies consistently in suspicious activity report narratives and add FIN-2024-DEEPFAKEFRAUD where media was manipulated.
- Share confirmed synthetic and theft indicators with consortium partners where your legal basis allows.
Standards and references
- Federal Reserve: Synthetic Identity Fraud
- Federal Reserve: Synthetic Identity Fraud in the U.S. Payment System (white paper)
- FinCEN: Identity-Related Suspicious Activity, 2021 Threats and Trends
- FinCEN: Alert on Fraud Schemes Involving Deepfake Media (FIN-2024-Alert004)
- NIST: SP 800-63A-4 Identity Proofing and Enrollment
- eCFR: 16 CFR Part 681 Identity Theft Rules
- US Code: 15 U.S.C. 1681c-2 Block of Information Resulting from Identity Theft
- FTC: New FTC Data Show a Big Jump in Reported Losses to Fraud
Related NHI Mgmt Group resources: Account Takeover vs New Account Fraud, Identity Proofing and KYC Guide, Fraud Data Sharing and Consortium Intelligence Guide, Merchant Fraud and Bust-Out.