Any online form that can authorise or save a payment card can be turned into a free validation service for criminals holding bulk stolen card data. Card testing, also called carding, enumeration or account testing, is the use of bots to submit many small or zero-value authorisations so that attackers learn which cards are live before they cash them out elsewhere. The merchant whose checkout is abused pays in fees, disputes, damaged authorisation rates and, increasingly, card scheme monitoring. This guide goes deeper than the overview in the Payment Fraud Guide: it covers the attack variants and the endpoints they hit, detection signals, a layered control set with example velocity rules, a phased rollout, metrics and the mistakes that let attacks continue. It is written for e-commerce engineers, payments and fraud teams, charities and platforms that take card payments online.
Key takeaways
- Every card endpoint is a target, not only checkout. Attackers prefer card-save and card-update flows because validation there often does not appear on a cardholder statement, and donation pages because $1 to $5 payments look normal.
- No single heuristic stops it. Blocking IP addresses alone fails against residential proxies; layer bot defence, velocity limits by several keys, issuer checks (CVC and postcode) and friction such as CAPTCHA or 3-D Secure under attack.
- Decline spikes are the main signal. A sudden rise in failed authorisations, generic declines and low-value payments with nonsense names is how most attacks show up first.
- Consequences outlast the attack. A high decline rate can make issuers treat all your traffic as riskier, and Visa's Acquirer Monitoring Program, in effect since 1 April 2025, includes enumeration criteria.
- Protect the keys and merchant credentials. Leaked payment API secret keys and compromised merchant IDs let attackers submit authorisations directly, bypassing every front-end control.
How card testing works
The PCI Security Standards Council and the National Cyber-Forensics and Training Alliance (NCFTA) describe account testing, which they also call payment account enumeration, card testing and BIN attacks, as a criminal testing payment account numbers to validate cardholder information in order to commit fraud. OWASP classifies the activity in its Automated Threats to Web Applications project, which separates validation from the later use of the cards.
| OWASP ID | Name | What the attacker wants | Where it hits you |
|---|---|---|---|
| OAT-001 | Carding | Confirm which cards in a bulk stolen set are valid, through many payment authorisation attempts | Checkout, card-save, card-update, donation and free-trial forms |
| OAT-010 | Card Cracking | Find the missing expiry date or security code for stolen card numbers | Any form that returns a different response for a wrong expiry or CVC |
| OAT-012 | Cashing Out | Buy goods or obtain cash with validated cards | Resaleable goods, gift cards, digital goods, often at a different merchant |
| OAT-019 | Account Creation | Mass-create accounts to get past controls that only apply to guests | Sign-up and customer creation APIs |
Variants named by PCI SSC and NCFTA
- BIN attacks: attackers take a known issuer prefix and use card number generators to fill in the remaining digits, then use your site to discover which generated numbers are live.
- CC checkers: tools, some free and some sold on criminal markets, that submit bulk card numbers with expiry dates and security codes against merchant sites.
- Carding across merchants: bots try one card with different expiry dates, postcodes and CVV values across many merchants at once, which defeats per-site attempt limits. Sites that do not check the security code make this easier.
- Low-value charges: charity and donation sites are common targets because they process many small payments, often $1 to $5.
- Compromised merchant IDs: criminals take over a merchant's ID and gateway credentials, often through weak or default passwords, and submit large volumes of low-value authorisations directly.
- Blind brute force: sequential numbers with no cardholder data; even a small success rate yields many valid cards at scale.
Stripe's documentation adds that card setup (saving a card without charging it) is a method attackers prefer because the validation does not usually show on a cardholder's statement, and that testers collect 3-D Secure or issuer responses to decide which cards are valid before cashing them out or reselling them.
Why it costs the abused merchant
| Consequence | Mechanism | Who notices first |
|---|---|---|
| Disputes and early fraud warnings | Some test payments succeed and cardholders report them | Payments operations, weeks later |
| Higher declines for real customers | Issuers associate your merchant with high decline rates and harden their stance, even after the attack ends | Conversion and revenue teams |
| Fees | Authorisation fees on custom pricing plans and dispute fees | Finance |
| Scheme monitoring | Visa's Acquirer Monitoring Program (VAMP) consolidated five fraud and dispute programmes and assesses enumeration by the number and rate of enumerated authorisations | Your acquirer |
| Infrastructure strain | Bot traffic load on checkout and payment APIs | Site reliability |
| Polluted business data | Test transactions look like new customers in revenue and growth figures | Analytics and marketing |
Detection signals
Most attacks are obvious once someone looks at the right chart. Combine the indicators from the PCI SSC and NCFTA bulletin with those from processor documentation:
- Decline spikes: a sudden rise in failed or blocked authorisations, in HTTP 402 responses from the payment API and in generic decline codes.
- Retry patterns: the same card number tried with changing expiry dates, security codes or postcodes, which is the signature of card cracking.
- Card number patterns: numbers from BIN ranges that were never issued, and sequential or regularly incremented numbers within one BIN.
- Odd payment content: low amounts with nonsensical customer names and email addresses, and more distinct cardholder names than normal from one source.
- Authorisations that never settle: a rise in approved authorisations that are never captured.
- Dormant or new accounts waking up: sudden volume at new, low-settlement or dormant merchant accounts, a sign of compromised merchant credentials.
- Account-level signals: many cards added to one account, many customers created from one IP address, and one card used from many IP addresses.
Feed these into your SIEM or fraud platform as alerts with thresholds relative to your normal baseline, for example declines per minute at three times the trailing seven-day average. The Fraud Metrics and KPIs Guide covers how to baseline them.
Layered controls
Reduce exposure
- Use the processor's recommended hosted fields or checkout where you can. Stripe, for example, states that its recommended integrations apply rate limiters, models and CAPTCHA triggers during attacks, and that their success depends on the risk data your integration sends (IP address, email, name, billing address, device signals).
- Require a session or login before card entry on flows that do not need guest access, and apply anti-CSRF tokens so scripts cannot post directly to the payment endpoint.
- Do not reveal which field was wrong. Return the same generic error for a wrong expiry, CVC or postcode, so card cracking gets no signal.
Throttle with velocity rules
Velocity limits must use several keys at once, because attackers rotate IP addresses and devices. Stripe's Radar rules guide gives examples such as blocking an IP address with more than one successful authorisation in the past hour, and blocking a card number with more than one blocked payment in the past hour. Tune numbers to your own baseline.
| Key | Example limit | Catches | Watch out for |
|---|---|---|---|
| IP address | Authorisations or blocked payments per hour; customers created per IP per day | Unsophisticated scripts | Shared corporate and mobile carrier IPs; rotating residential proxies |
| Card number or card fingerprint | Attempts per card per hour, across all accounts | Card cracking (OAT-010) | Genuine customers retyping a CVC: allow two or three tries |
| Account or customer | Cards added per account per day | Card-save testing | Families sharing an account |
| Device or browser fingerprint | Distinct cards per device per hour | Distributed attacks behind proxies | Privacy controls and fingerprint drift |
| BIN range | Attempts or declines per BIN per minute | BIN attacks | Large issuers with many genuine customers |
| Product or amount | Purchases of the same low-value item or donation amount per hour | Low-value testing on donation pages | Campaign peaks and appeals |
| Endpoint | Lower limits on zero-value, card-save and card-update requests | Card setup testing | Subscription renewals with legitimate retries |
Add friction under attack
- CAPTCHA: validate it server side on every request that can validate a card or make a payment, not only on page load. If attacks continue, raise the score threshold or switch from invisible to visible challenges. A useful pattern is to let the first attempt from an IP address through and require a challenge for further attempts in the next few hours.
- 3-D Secure: step up to EMV 3-D Secure for risky sessions or for all transactions during an attack. Note that attackers also read 3DS responses, so pair it with velocity limits.
- Issuer checks: require the security code and postcode, and decline when the issuer does not confirm them.
- Random delays and temporary blocks: the PCI SSC and NCFTA bulletin recommends random pauses during card checks to slow brute force, and lockouts after a set number of failures.
Bot defence at the edge, through a bot management service or web application firewall rules, removes much of the traffic before it reaches the payment API. See the Bot Management Guide and the Web Application Firewall Guide.
Where identity fits
Card testing is an abuse of non-human credentials as often as it is a bot problem. Stripe warns that a leaked secret API key lets testers create payments and save cards directly, and the PCI SSC and NCFTA bulletin describes criminals taking over merchant IDs and gateway portal credentials. Keep secret keys server side only, use restricted keys scoped to the operations each service needs, rotate keys after any suspected exposure, require strong unique passwords and MFA on gateway portals, and alert on authorisations from merchant accounts or API keys that are normally idle. Card-on-file accounts are also targets for credential stuffing, so the same login protections apply. See the API Key Management Guide and the Password Spraying and Credential Stuffing Defence Guide.
Decision guidance
- Hosted versus direct integration: small merchants and charities should use the processor's hosted checkout or fields, which bring built-in testing defences. Direct API integrations give more control but you then own CAPTCHA, velocity and data enrichment.
- Visible versus invisible challenge: invisible challenges protect conversion in normal times; switch to visible challenges or mandatory 3DS only while an attack runs, then step back down.
- Block versus slow: outright blocks on IP or BIN are blunt and easy to evade. Slowing (delays, step-up) costs the attacker more and harms fewer genuine customers.
- Small versus large organisations: a small shop can rely on processor tooling, a decline alert and a playbook. A large retailer or platform needs its own velocity service across all card endpoints, fingerprinting, and a direct line to its acquirer's fraud team.
How to implement
- Map every card endpoint: checkout, card-save, card-update, subscription sign-up, donation, free trial and any partner or mobile API. Attackers find the weakest one.
- Baseline: record normal declines per minute, approval rate, cards per account and payments per IP for at least two weeks.
- Alert first: set decline-spike and low-value-burst alerts to an on-call owner before adding blocks.
- Deploy controls in monitor mode: run velocity rules and challenge scoring in log-only or review mode, and measure how many genuine customers would have been affected.
- Enforce in rings: enable on the lowest-risk endpoint (often card-save) first, then donation and checkout, watching approval rate and complaints.
- Write the attack playbook: who can switch on mandatory challenges or 3DS, who calls the acquirer, how to refund successful test payments to head off disputes, and how to stop subscription retries on fraudulent customers.
- Roll back cleanly: keep each emergency control behind a feature flag with an owner and an expiry, so friction is removed when the attack ends.
- Review after every attack: record the endpoint, the keys that would have caught it earlier and the rule changes made.
Detection and measurement
| Metric | Why it matters |
|---|---|
| Authorisation decline rate by endpoint and BIN | The primary attack signal and what issuers and schemes see |
| Time from attack start to mitigation | Shows whether alerting and the playbook work |
| Share of attack attempts stopped before authorisation | Measures edge and velocity controls, since attempts that reach the issuer still count against you |
| Successful test payments refunded before dispute | Limits dispute ratios and fees |
| Genuine customer challenge or false decline rate | Keeps friction honest |
| Authorisations from idle API keys or merchant IDs | Detects credential compromise |
Framework mapping
| Control area | NIST CSF 2.0 | NIST SP 800-53 Rev. 5 | OWASP ASVS 5.0 |
|---|---|---|---|
| Anti-automation on payment functions | PR.PS (platform security) | SC-5 Denial-of-service Protection | V2.4.1 anti-automation; V2.4.2 realistic human timing |
| Decline and velocity monitoring | DE.CM-09, DE.AE-02 | SI-4 System Monitoring | V16.3.3 logging attempts to bypass controls |
| Correlating processor, edge and app data | DE.AE-03 | SI-4 | V15.3.4 preserving the original client IP |
| Lockout and retry limits | PR.AA | AC-7 Unsuccessful Logon Attempts | V6.1.1 documented rate limiting and anti-automation |
| Containing an active attack | RS.MI-01 | IR-4 Incident Handling | Not applicable |
In OWASP's automated threat taxonomy the activity is OAT-001, OAT-010 and OAT-012. PCI DSS v4.0.1 compliance gives a baseline for protecting card data and merchant credentials, but it does not on its own stop your site being used as a validator.
Common failure patterns
- Protecting checkout only: attackers move to the card-update page or a free-trial sign-up that runs a zero-value authorisation with no CAPTCHA.
- Client-side CAPTCHA: a challenge rendered in the browser but never verified on the server is ignored by scripts that post straight to the API.
- Leaking the validation result: distinct error messages for a wrong CVC versus a wrong expiry give card cracking the answer it needs.
- Retrying fraudulent cards: automatic subscription retries on cards saved by attackers repeat the attack against issuers; stop retries for customers flagged after an attack.
- Exposed keys and portals: a secret key in client code or a gateway portal with a default password lets testers bypass every front-end control.
- No owner out of hours: attacks often run overnight; without on-call alerting, thousands of authorisations go through before anyone sees the decline chart.
Lessons from incidents
Card testing is the step between mass card theft and cash-out. The AI Agent Retail Card Theft Campaign 2026 shows how quickly bulk card records can be gathered: more than 600,000 unexpired card records were taken from two companies, with checkout skimmers on many more sites. The incident write-up does not describe how the cards were validated, but stolen sets of that size are exactly what card testing is used to sort, and merchants with open, unthrottled card endpoints are the ones that end up doing the sorting.
Practitioner checklist
- List every endpoint that can authorise, verify or save a card, including APIs and mobile apps.
- Use the processor's recommended hosted integration where possible and send full risk data with each payment.
- Verify CAPTCHA server side on all card validation and payment requests.
- Apply velocity limits by IP, card, account, device, BIN and endpoint, tested in monitor mode first.
- Return one generic error for any card detail mismatch and require CVC and postcode checks.
- Alert on decline spikes and low-value bursts to an on-call owner, day and night.
- Keep payment secret keys server side, restricted, rotated and monitored; protect gateway portals with MFA.
- Maintain an attack playbook covering emergency friction, acquirer contact, refunds and stopping retries.
- Review decline rates by endpoint and BIN monthly with your acquirer's monitoring in mind.
Standards and references
- PCI SSC and NCFTA: Account Testing Bulletin
- OWASP: Automated Threats to Web Applications
- OWASP: OAT-001 Carding
- OWASP: Application Security Verification Standard 5.0
- Visa: Introducing the Visa Acquirer Monitoring Program
- Stripe: Protect yourself from card testing
- Stripe: Radar rules 101
- NIST: SP 800-53 Rev. 5
Related NHI Mgmt Group resources: Payment Fraud Guide · Bot Management Guide · Card Payments Security for Small Merchants Guide · Loyalty Points and Gift Card Fraud Guide