Back to Blog
DDoSDDoS TestingBankingFraudOnline Banking

DDoS Testing for Banks: Customer Access, the Core and Fraud Behind the Flood

BlackNeuron Research Team
November 22, 2026
14 min read
DDoS Testing for Banks: Customer Access, the Core and Fraud Behind the Flood

When a bank is under DDoS, its customers ask one question within minutes: is my money still there?

The attack cannot touch the money. It can stop people seeing it, stop them freezing a stolen card, and keep the bank's own staff from watching a fraudulent transfer leave while everyone is looking at the flood.

That makes a bank's DDoS test different from a load test on a busy website. The pages are the easy part. The hard parts sit behind them and beside them.

DDoS testing for banks, defined

DDoS testing for banks checks two things at once, with permission and under a flood: that customers can still reach and control their accounts, and that the bank can still see what is happening to them. In practice that means online and mobile banking login, the core banking calls behind every logged-in page, the authentication channel that also confirms payments, the contact centre that takes the overflow, and the fraud operation that keeps working while the incident runs.

The rails themselves, and the clocks that card schemes and payment schemes hold over a bank, have their own post on DDoS testing for financial services. Approvals, ramps and abort criteria are no different for a bank than for anyone else, so use the complete guide to DDoS testing for those. This page stays on the retail side: the customer, the core and the fraud desk.

PathWhat the flood actually threatensWhat the test measures
Logged-in pagesThe core banking connection poolCore call latency and pool occupancy
Login and step-upThe one-time-code channel shared with payment approvalCodes delivered per second, end to end
Contact centreThe same back end, reached by phoneQueue length, time to block a card
Fraud operationsAttention, alert queues and held paymentsAlert latency on planted test transactions
Hosted online bankingA platform shared with other banksThe provider's evidence, your own fallbacks

The campaign that taught banks what attackers aim at

From September 2012 into 2013, a campaign that called itself Operation Ababil ran sustained floods against the online banking sites of dozens of US banks. In March 2016 the US Department of Justice charged seven people over attacks on 46 financial institutions across 176 days (reported by Reuters, March 2016).

Three details from that campaign still shape how banks should test.

The targets were the customer-facing sites

The floods went for online banking, not for card switches or payment systems. Customers could not log in or see balances, and that was the outcome the campaign wanted: visible, embarrassing, and repeated.

The botnet ran on servers, not home routers

The traffic came from a botnet of compromised web servers, known as Brobot, sitting in hosting data centres. Each source had far more bandwidth than a home device and could send a heavy stream of HTTP requests on its own.

A per-IP rate limit tuned for millions of small sources behaves very differently against a few thousand large ones. A botnet built on hosting servers is worth modelling separately from an IoT one.

The targets were announced in advance

The group posted the next week's targets publicly. Banks knew roughly when their turn would come.

That turns readiness into something testable on a calendar. Can the bank raise its posture for a named day: pre-provision scrubbing, lower thresholds, staff the right desks, warn the contact centre? A ransom DDoS note with a deadline poses the same question.

Online banking is a thin layer over an old core

The web and app front ends are modern, scaled out and usually behind a CDN. The thing they talk to is often not.

The channel answers, the core decides

Between the front end and the ledger sits channel middleware: an API gateway or integration layer that turns a page view into calls on the core banking system. That layer holds a fixed pool of connections to the core, and the core itself has an online transaction capacity sized for branches plus normal digital traffic.

Scaling the web tier does nothing for either. A flood that gets past the edge with valid-looking requests is a flood on the core, filtered through a pool that was never sized for it.

Four channels, one pool, one core The front ends scale out. The connections behind them do not. Web banking flood arrives here Mobile app same API calls IVR self-service balance, card block Agent desktop account lookups Channel middleware API gateway or integration layer fixed pool to the core 80 connections core calls Core banking system online capacity sized for branches plus normal digital traffic Nightly batch window snapshot balances, memo posting fewer online connections a flood here meets a smaller path Scaling the web tier adds no core connections. The pool and the batch window set the ceiling. BlackNeuron™
DDoS testing for banks: web, mobile app, IVR and contact centre agent desktop all call the same channel middleware, which holds a fixed pool of connections to the core banking system, so a flood on any channel competes for the same core capacity, and the core's online capacity shrinks during the nightly batch window

Every logged-in page is a core call

The landing page after login shows balances and recent transactions. Those cannot be cached for long, because a customer who has just been paid expects to see it.

So an attacker who can get through login, or who floods the endpoints a logged-in session calls, is buying core capacity with every request. The cost per request is not visible at the edge; API floods show the same asymmetry, and in a bank the expensive endpoint is usually the one that reads the ledger.

The batch window shrinks the path

Many core systems still run an end-of-day batch. While it runs, online channels are served from a reduced path: a snapshot balance, memo posting of new transactions, fewer online connections, slower responses.

A flood that lands in that window meets a smaller pipe. If it slows the batch itself, the damage spreads past the online channel into the next morning's postings and into the settlement cut-offs that the financial services post covers.

The arithmetic of a slower core

Connections held open equal calls per second multiplied by seconds per call; that is Little's law. Take an illustrative 250 core calls a second at 20 milliseconds each: 5 connections in use, against a middleware pool of 80.

Now let core latency climb to 800 milliseconds, through batch, through a flood, or both. The same 250 calls a second want 200 connections. The pool pins at 80, new calls queue, and the queue is what customers see as a login that spins and a balance that never loads. No server fell over.

2026-11-22T07:08:10.339616 image/svg+xml Matplotlib v3.5.1, https://matplotlib.org/ 0 20 40 60 80 100 120 140 Time (seconds) 0 20 40 60 80 Connections in use pool cap: 80 core slows 250 core calls a second; core latency climbs from 20 ms to 800 ms; middleware pool of 80. Model, not a measurement. Core connections pin while every server stays up connections to the core BlackNeuron™
Connections from channel middleware to the core banking system rising as core latency climbs from 20 to 800 milliseconds at a steady 250 calls a second, until the count pins at the pool cap of 80 and further calls queue

Find the rate at which the pool pins

Measure core call latency and middleware pool occupancy, not page load time at the edge. Run the profile twice: once in daytime conditions and once in the batch-window conditions, the second in a production-like environment if the live core cannot take it.

Record the request rate at which the pool pins. That number, not the CDN's capacity, is the online bank's real ceiling.

The code that logs you in also approves your payments

Login floods at banks rarely stop at the password form, and the reason is specific to banking.

Credential floods walk the whole path

Attackers hold large lists of leaked username and password pairs. Sending those at volume makes the flood indistinguishable from customers who mistyped, and a fraction of the pairs work, so the requests pass the password check and reach step-up authentication.

The account lockout side of this, where your own policy locks real customers out, is covered in the post on VPN gateway DDoS testing. In a bank, a locked-out customer phones, which feeds the next section.

Step-up shares a channel with payment approval

Under PSD2's strong customer authentication rules, a payment code must be dynamically linked to the amount and the payee (Article 5 of Delegated Regulation (EU) 2018/389). Many banks deliver both login codes and payment approval codes through the same channel: an SMS aggregator, a push service, or an authenticator back end.

That channel has a throughput limit. Messages per second per sender, push requests per second per app, signatures per second on the back end.

A login flood that exhausts it does not only stop logins. It stops customers who are already logged in from approving a transfer, because their approval code is queued behind the attacker's login codes.

The login flood spends the payment channel Login codes and payment approval codes often leave through the same gateway. Attack logins leaked credential pairs Password check a share of pairs pass Customer payments already logged in approval code, linked to amount and payee Code channel SMS, push or authenticator fixed messages per second Login codes fill the queue first Approval code late payment times out Test it on test accounts: codes delivered per second and approval delivery time, measured together. BlackNeuron™
DDoS testing for banks, the shared authentication channel: attacker login attempts with leaked credentials that pass the password check trigger one-time codes, and customer payment approvals need codes from the same SMS or push channel, which has a fixed messages-per-second limit, so approval codes queue behind attack traffic and payments time out

The channel is also a bill

Every SMS sent is charged, and attackers who trigger codes to premium numbers turn the flood into revenue for themselves. That metering side belongs to EDoS testing; for this test the point is capacity.

Prove whether the channels are shared

Use test accounts with test phone numbers or a provider sandbox, never real customers. Flood the login path with test credentials that pass the password check, and measure two numbers together: login codes delivered per second, and the time for a payment approval code to arrive on a separate test account. If the second rises with the first, the channels are shared, whatever the architecture diagram says.

When the app fails, customers phone

A bank's contact centre is its overflow channel, and it is rarely tested as part of a DDoS exercise.

The phone line reaches the same back end

The IVR's self-service options (balance, recent transactions, card block) call the same middleware the app calls. The agent desktop does too. So a flood that pins the core pool also slows the menu, and slows every agent looking up a caller's account.

Call volume rises at the moment each call takes longer to handle. Queue length grows on both counts.

Freezing a card is a safety control

The in-app card freeze goes down with the app. The lost and stolen card line is then the only way a customer can stop a card, and that is a control the bank should treat like any other security control: it must work during the incident, not after it.

Check where the card block actually executes. If it runs through the same channel middleware, the hotline inherits the flood.

Put the phones in the test window

During a test window, with the contact centre briefed, have a small team place scripted calls on test accounts. Measure time to answer, IVR completion rate, agent lookup time and the time from call start to a confirmed card block. Repeat with the middleware pool pinned. Those figures belong in the report beside the request rates.

A flood can be cover for fraud

The flood is often not the point. It is noise.

Regulators have said so plainly

US banking regulators' joint FFIEC statement on DDoS attacks (April 2014) warned that some attacks had been used to divert attention while fraud was carried out, and asked institutions to keep that possibility inside their incident response. It is the reason a bank's DDoS plan cannot be owned by the network team alone.

What the incident takes away

A large flood draws the most experienced people into a bridge call. It fills the security monitoring pipeline with flood logs, so fraud and account-takeover alerts wait behind them. Payment holds pile up as risk checks slow, and someone senior asks whether the hold limits can be relaxed to clear the backlog.

Every one of those is the window a fraud crew wants. Account takeover attempts and high-value transfers go out while the fraud analysts are short-staffed and their alerts are late.

The flood is the noise, the transfer is the point What an incident takes away from the fraud desk, minute by minute. Flood Bridge call Monitoring Payment holds Fraud crew flood at the edge, everyone watching it senior staff pulled onto the call ingest full of flood logs, fraud alerts wait behind them holds build up, pressure to relax the limits takeover, large transfers 0 min 10 20 30 40 the window a fraud crew wants illustrative timing BlackNeuron™
Timeline of a DDoS used as cover at a bank: the flood starts, a bridge call pulls senior staff, monitoring ingest fills with flood logs and fraud alerts queue behind them, payment holds build up and pressure grows to relax them, while account takeover and high-value transfers run in parallel during the gap

What the test proves

Make the fraud operation part of the test, not a bystander. While the flood runs, push a small set of planted transactions through test accounts that should trip the fraud rules: a new payee and a large transfer, a login from an unusual device followed by a change of phone number.

Then measure what happens to them. How long until the alert fires, how long until an analyst sees it, and whether the payment is held. Check also that nobody changed a velocity limit or released held payments to clear a queue during the run. A DDoS war game is the place to rehearse that pressure without traffic first.

Keep fraud monitoring off the flood's path

Where the fraud engine shares log ingest, message queues or network paths with the internet edge, give it its own. An engine that answered every request but raised its alert twenty minutes late has failed this test.

Many banks share one online banking platform

Smaller banks and credit unions often do not run their own online banking. A hosted digital banking provider runs it for them, and for many other institutions on the same infrastructure.

Your login page is a tenant

The bank's online banking hostname points at the provider. An attack aimed at any bank on that platform lands on the shared front door, and every bank on it can feel it.

How far that reaches depends on how the provider isolates tenants, and the mechanics (silo or pooled, per-tenant limits, cells) are set out in DDoS testing for SaaS platforms. From the bank's side the question is narrower: what happens to my customers when the target is someone else?

One bank is the target, every tenant feels it Smaller banks and credit unions often run online banking on a shared hosted platform. Bank A hostname the target Bank B hostname Credit union C Bank D hostname Hosted banking platform shared edge shared login and app tier shared connectors to cores Flood for Bank A lands here on capacity every tenant uses Bank A core customers locked out Bank B core customers slowed Credit union C core customers slowed Bank D core customers slowed Loading this platform needs the provider's written permission, and the load reaches the other banks. BlackNeuron™
Several banks' online banking hostnames point at one hosted digital banking platform, which connects back to each bank's own core; a flood aimed at one bank's hostname hits the shared edge and platform, so customers of the other banks on the same platform are slowed too

The right to test belongs to the provider

You cannot load a shared platform without the provider's written permission, and even with it you would be loading other banks' customers. So for a hosted bank the test evidence comes partly from the provider.

Ask for it in writing: how a targeted tenant is isolated, whether its hostname can be moved or rate-limited without affecting the rest, the provider's recovery objective for the platform, and when it last tested any of that.

What a hosted bank can still test itself

Plenty. Your own DNS and any edge you put in front of the hostname. Your contact centre, card block line and status page. Your own core, if you run it. And the scenario where the platform is down for everyone: who decides what customers are told, and through which channel. That scenario needs no traffic at all; a tabletop session with the provider's contact in the room covers it.

Telling customers, from somewhere the flood cannot reach

A bank's status message is part of the security response, because silence reads as "something is wrong with my money".

Host the status page on separate infrastructure and a separate provider from online banking. Prepare wording that separates "you cannot log in" from anything about funds, and agree who can publish it in minutes rather than hours. Phishing crews read outages too, and a clear official channel is what customers check before they trust a text claiming to be from the bank.

What supervisors expect

In the US, the 2014 FFIEC statement asked institutions to assess DDoS risk, plan with their internet service and hosting providers, and make sure incident response covers a sustained attack. British supervisors frame the question as impact tolerances for important business services, which the post on UK DDoS testing works through. EU banks fall under DORA (DORA DDoS testing), and threat-led work for UK firms runs through CBEST.

This is orientation, not legal advice. What examiners tend to ask for, under any of these, is evidence per customer path: the planted fraud transfer that was held, the card block that went through, the pool that pinned at a known rate and was resized.

A bank DDoS test plan, path by path

PathBefore the testDuring the testPass looks like
Core behind the channelsFind the middleware pool size and the batch windowMeasure core latency and pool occupancy, day and batch profilesPool does not pin below the agreed attack rate
Login and codesSet up test accounts, test numbers or a provider sandboxLogin codes per second against approval code delivery timeApproval codes arrive inside their timeout
Contact centreBrief the floor, script test callsTime to answer, IVR completion, time to block a cardCard block confirmed within the target
Fraud operationsWrite planted transactions on test accountsAlert latency, analyst pickup, payment held or notAlerts on time, no limits relaxed
Hosted platformGet the provider's written evidence and permissionTest only what is yours: DNS, edge, phones, status pageFallbacks work with the platform degraded
CommunicationStatus page off the main stack, wording agreedPublish a test notice inside the target timeCustomers have an official channel

FAQ

Why do attackers go after online banking rather than payment systems?

Online banking faces the internet by design, and an outage there is public. Payment systems sit on private networks with the schemes. The best-known campaign against US banks, Operation Ababil in 2012 and 2013, went almost entirely for customer-facing sites.

Does a DDoS attack put customers' money at risk?

Not directly. A flood does not move money. The risk is what runs under its cover: account takeover and fraudulent transfers while fraud analysts are short-handed and their alerts are delayed, and customers unable to freeze a stolen card.

Can a credit union test its DDoS resilience if online banking is hosted?

Yes, within limits. The platform belongs to the provider, so loading it needs the provider's written permission and affects other institutions. The credit union can test its own DNS, contact centre, card block line, status page and communications, and should ask the provider for its own isolation and recovery evidence.

Should one-time passcodes be part of a bank DDoS test?

Yes. A login flood with credentials that pass the password check triggers codes, and the same channel often delivers payment approval codes. Measure both together on test accounts, never on real customers.

Does a bank DDoS test need to cover card and payment rails too?

A complete one does, but those paths are tested against scheme simulators and their timeouts, which our financial services post covers. This page adds the customer's side: whether people can get into their accounts and whether the fraud team can see what moves while they cannot.

Two kinds of sight

A bank in a DDoS has two audiences watching the same accounts. Customers want to see their money and stop it moving. The fraud team wants to see it moving when it should not.

A test that only counts requests at the edge measures neither. Can a customer load a balance, approve a payment and freeze a card in the twentieth minute of the flood? Did a planted fraudulent transfer get held while it ran?

Ask both questions in the same window. The attacker already does.