A 2018 hacking complaint showed thieves editing the bank’s own record of its payments

The September 2018 charges over the Bangladesh Bank theft described intruders who deleted, intercepted and altered the records staff used to check what was sent. A bank needs a payment record its gateway cannot change.

Looking Back revisits past weeks one at a time, up to the most recent, and asks what it meant for the banks, health systems and public agencies we serve. The series was written in 2026, with the benefit of hindsight, and each article says which week it looks back on.

Two-minute summaryThe dek and the first sentence of each section, quoted from the article

The September 2018 charges over the Bangladesh Bank theft described intruders who deleted, intercepted and altered the records staff used to check what was sent. A bank needs a payment record its gateway cannot change.

  1. The week On Thursday, September 6, 2018, the Department of Justice unsealed a criminal complaint charging a North Korean citizen, a computer programmer, with conspiracy to commit computer fraud and abuse and conspiracy to commit wire fraud.
  2. The way in: a mailbox and a long wait According to the complaint, the targeting of banks in Bangladesh began as early as October 7 and 8, 2014, with online reconnaissance of specific banks the subjects later sent spear-phishing messages to.
  3. The sending endpoint: authenticated messages that were not authorized The complaint describes Bangladesh Bank’s SWIFTLIVE system as the core component of its SWIFT processing environment.
  4. The bank’s own record: what the staff were shown The complaint’s central allegations for banks concern records.
  5. The money’s path: correspondents, beneficiaries and screening The complaint says the fraudulent messages directed the Federal Reserve Bank of New York to transfer funds from Bangladesh Bank’s U.S. dollar account there to specified accounts in the Philippines and Sri Lanka, via specific U.S. correspondent banks.
  6. Signals the complaint puts on the record The complaint is also a catalog of signals that a bank’s monitoring could look for.
  7. Designing the independent record The recommendation at the center of this article, a record the sending endpoint cannot change, can be met in more than one way.
  8. How the four parts cover for each other The value of separating the four parts is that a bank can see where one part’s failure is caught by another.
  9. What the unsealing gave defenders The Justice Department’s release described the mitigation work that ran alongside the case.
  10. Objections One objection is that a state-backed team with years of patience will get in anywhere.
  11. How it applies across institutions A large bank may run its own gateway with a dedicated team.
  12. With hindsight North Korean government hackers were soon visible in a different payment system.

The week

On Thursday, September 6, 2018, the Department of Justice unsealed a criminal complaint charging a North Korean citizen, a computer programmer, with conspiracy to commit computer fraud and abuse and conspiracy to commit wire fraud. The complaint had been filed on June 8, 2018 in federal court in Los Angeles. The department said the two charges carried maximum sentences of five years and 20 years in prison, and that the charges were accusations, with the defendant presumed innocent unless proven guilty.

The complaint alleged that the programmer was a member of a government-sponsored hacking team that the private sector called the Lazarus Group, and that he worked for Chosun Expo Joint Venture, a North Korean government front company with offices in China and North Korea that the department said was affiliated with Lab 110, a component of North Korean military intelligence. The department tied the conspiracy to the creation of the malware used in the WannaCry 2.0 ransomware attack of May 2017, the theft of $81 million from Bangladesh Bank in 2016, the attack on Sony Pictures Entertainment in November 2014 and intrusions or attempts against defense contractors, financial firms, technology and virtual currency companies, academia and electric utilities.

On the financial side, the department’s summary described the Bangladesh Bank theft in a few lines. The conspirators compromised the bank’s network with spear-phishing emails, reached the computer terminals that interfaced with SWIFT, the interbank messaging network, and sent fraudulently authenticated SWIFT messages directing the Federal Reserve Bank of New York to transfer funds from Bangladesh Bank to accounts in other Asian countries. The department said the conspiracy attempted to gain, and did gain, access to several other banks in various countries from 2015 through 2018 using similar methods and watering hole attacks, attempting the theft of at least $1 billion through such operations.

The same day, the Treasury Department’s Office of Foreign Assets Control designated the programmer and Chosun Expo Joint Venture. Treasury designated the programmer under Executive Order 13722 for engaging in significant activities undermining cybersecurity on behalf of the government of North Korea or the Workers’ Party of Korea, and designated the company as an agency, instrumentality or controlled entity of the government of North Korea. Treasury said that, as a result, any property or interests in property of the designated persons in the possession or control of U.S. persons or within the United States must be blocked, and U.S. persons were generally prohibited from dealing with them.

The complaint, as posted with its attachments, ran to 179 pages. It explained how the FBI linked the attacks: through email and social media accounts used for spear-phishing, aliases, accounts used to store stolen credentials, shared malware code, proxy services and North Korean and Chinese IP addresses. For banks, its central pages were the forensic account of what the intruders did inside Bangladesh Bank and other banks, and in particular how they handled the banks’ own records of the messages they had sent. That account is the subject of this article. The complaint’s allegations had not been tested in court.

The headline of the complaint was attribution: the United States naming a North Korean programmer for WannaCry, Sony and a central bank theft. For a bank’s technology and security leaders, the more durable material is in the forensic pages about Bangladesh Bank and the other banks the complaint describes. They show intruders who did not stop at sending fraudulent payment messages. They also changed what the bank’s own systems recorded about those messages, so that the people who checked the records would see nothing wrong.

Our argument is that the complaint described an attack on four separate things a bank relies on, and that a bank’s defenses should be organized the same way. The way in: the email and workstations through which the intruders entered and waited. The sending endpoint: the system that creates and authenticates payment messages. The bank’s own record: the database, printouts and documents that tell staff what was sent. The money’s path: the correspondent banks and beneficiary accounts that move funds after a message leaves. The four do not overlap. Each is controlled by different people, and a failure in one can be caught in another only if the bank has designed it that way.

We take each in turn, drawing on the complaint’s allegations, which have not been tested at trial. We then test the argument against objections and look at how it applies to banks of different sizes, to other payment systems and to institutions outside banking. Institutions should take advice from counsel on any legal obligations that apply to them, including sanctions obligations.

The way in: a mailbox and a long wait

Reconnaissance and spear-phishing

According to the complaint, the targeting of banks in Bangladesh began as early as October 7 and 8, 2014, with online reconnaissance of specific banks the subjects later sent spear-phishing messages to. A subject researched Bangladesh Bank in February 2015 and another bank in Bangladesh in October 2015. The complaint describes spear-phishing emails offering a resume and cover letter, with a link to download a file. On August 11 and 12, 2015, one account sent twenty-five such messages to employees of multiple Bangladesh-based banks.

Forensic review after the theft, the complaint says, found that at least three Bangladesh Bank computers had attempted to download the file from one of those links between January 29 and February 24, 2015. The users of two of those computers matched addressees of a spear-phishing email, and the user of the third matched a contact in the address book of an account the subjects used.

Persistence that looked like normal traffic

In March 2015, the complaint says, the subjects moved within the bank’s network and saved a backdoor that communicated over a custom protocol designed to look like TLS, the encryption used for secure web traffic. The affiant explained that by imitating TLS, the traffic could pass intrusion detection systems that assume encrypted traffic cannot be read. The complaint describes this fake TLS technique as a common feature of Lazarus Group malware.

Then, on January 29, 2016, which the complaint describes as days before the fraudulent transfers, the subjects moved laterally through the network, including to the bank’s SWIFT processing environment. By the complaint’s account, the intruders had been in the bank’s network since early 2015 when they reached the payment system.

A regulator’s website as the lure

The complaint describes a second way in that bypassed the mailbox entirely. In January 2017, it says, the FBI learned of a campaign against the Polish banking sector. The intruders infected the website of the Polish Financial Supervision Authority, the country’s financial supervisor, and used it as a watering hole: a site the intended victims were expected to visit in the normal course of business. The malware on the site checked visitors against a list of IP address ranges and infected only selected ones, including, the complaint says, addresses assigned to banks. The complaint says the watering hole was likely in place from October 5, 2016 through February 2, 2017, and that the FBI had no evidence that any fraudulent transfers occurred in that incident.

For a bank, the watering hole is an uncomfortable precedent. Sites that staff trust, including a regulator’s, are sites an attacker who studies banks will choose. Allowing a site because it belongs to a regulator is a reasonable business decision, and it means that browsing protections cannot rest on the reputation of the site alone.

What this part asks of a bank

The way in is the part of the attack that resembles any other intrusion, and the defenses are familiar: filtering of email and links, restrictions on what workstations can run, and monitoring for unusual outbound connections. The complaint adds two points specific to banks. The first is the length of the wait: a bank that hunts for intruders only around payment systems will miss the long period they spend elsewhere. The second is the disguise: traffic that looks encrypted is not therefore safe, and a monitoring program that ignores what it cannot read is the program these intruders were built to pass.

The way in is owned by the bank’s enterprise security function. It is the widest of the four parts, and it cannot be made perfect. That is the reason the other three parts must not depend on it.

The sending endpoint: authenticated messages that were not authorized

The gateway

The complaint describes Bangladesh Bank’s SWIFTLIVE system as the core component of its SWIFT processing environment. It used the SWIFT Alliance Access application, which the complaint calls a customer-managed gateway to the SWIFT network that transmitted and received messages from other banks to create and confirm financial transactions. The complaint says the intruders made at least four attempts to log in to the computer hosting that system, and deleted some evidence of the attempts. One attempt used the name of a South American currency exchange that the same group had targeted, a name for which Bangladesh Bank had no account or credentials. The complaint treats that as a likely error by people running several intrusions at once.

Once in, the intruders could, in the complaint’s words, craft, authenticate and send SWIFT messages that appeared to be authentic and to originate from Bangladesh Bank’s own computer system. The complaint’s general description of the bank heists makes the same point: the messages were real but fraudulent, authenticated messages sent from the victim bank’s systems, which were being remotely accessed to construct them, but not actually authorized by the bank.

Authentication is not authorization

That distinction is the center of this part. The messaging network’s security assured the receiving bank that a message came from Bangladesh Bank’s gateway. It could not assure anyone that a person at Bangladesh Bank with authority to order the payment had decided to send it. Once the intruders controlled the gateway, the network’s authentication worked for them.

For a bank, that means the controls that prove authorization must sit around the gateway, not inside the messaging network. They include separating the gateway from the general network, requiring strong authentication for anyone who can log in to it, limiting which accounts can create and release messages, and checking messages against business records created elsewhere before release. The complaint’s account of four login attempts, one using the wrong name, shows that the gateway’s own login records can carry signals worth watching.

Who owns the endpoint

The sending endpoint is owned by the payments operation and the team that runs the gateway. Where it is a small team with a specialized system that the enterprise security function sees only from outside, we read the complaint as suggesting that the arrangement is a risk in itself: the intruders treated the gateway as one more host on a network they already controlled. The owner of the endpoint has to treat it as among the bank’s highest-risk systems, with its own access rules and monitoring, and the security function has to be able to see into it.

The bank’s own record: what the staff were shown

Deleting messages from the database

The complaint’s central allegations for banks concern records. It says that Alliance Access, as it received messages, recorded local copies of them, including by formatting and printing them to files or a printer, and by entering information about them in a separate database. The intruders interfered with each of those processes.

According to the complaint, a piece of malware named evtdiag.exe was configured to access the database that stored records of messages on Bangladesh Bank’s SWIFT server and was used to delete the specific messages that instructed the fraudulent transactions. It was designed to have another program securely delete it at 6:00 a.m. local time on February 6, 2016. Bank personnel shut the server down on February 5, and when it was started again on February 6 the deletion instruction was not sent, which the complaint describes as an apparently inadvertent preservation of the malware.

The complaint also says that at Bangladesh Bank, the intruders modified, in the computer’s memory, a copy of an Alliance Access file that would ordinarily prevent changes to the database that recorded the bank’s SWIFT messages. At a bank in Africa, it says, the same kind of modification was made to the file on the server’s disk.

Intercepting the printer

The paper record was attacked too. The complaint says that at Bangladesh Bank the intruders used malware named nroff.exe, the same name as a legitimate tool Alliance Access uses to format messages for printing, to intercept and modify fraudulent transactions that would otherwise have been automatically printed for the bank’s records. At the African bank, a file with that name appeared in the Alliance Access program folder on the day the unauthorized messages were sent and was erased later that day.

Altering the documents staff read

The complaint’s description of a bank in Vietnam shows the same idea adapted to a different process. That bank’s connection to SWIFT was managed by a third-party company. Each message sent to or from the bank was stored as a PDF document on the third party’s server, and bank employees connected remotely to that server and read the documents with a PDF reader. The malware recovered there was designed so that when employees opened those documents, it checked whether a document would show evidence of the fraudulent messages and, if so, modified it before the reader displayed it. The complaint says the concealment was tailored to unique aspects of that bank’s business processes.

What this part asks of a bank

In all three banks, the record that staff relied on to know what had been sent was produced by, or passed through, the same environment the intruders controlled. The database lived on the SWIFT server. The printouts were formatted by a program in the Alliance Access folder. The PDFs sat on a provider’s server and were opened through malware posing as the reader staff used. When the sending endpoint fell, the record fell with it.

Our central recommendation follows. A bank needs at least one record of what it sent that the sending endpoint cannot change. In practice, that can be a statement or confirmation from the counterparty, such as the correspondent bank that holds the account; an independent copy of message traffic captured outside the gateway’s host; or reconciliation of the bank’s nostro accounts against data from the correspondent, done by staff and systems that do not depend on the gateway. The detail matters less than the principle: the check has to read data the intruder could not edit.

Ownership of this part needs to be explicit. The record belongs to finance and reconciliation as much as to payments operations, and the independence of the record belongs to the security architecture. The complaint gives a board a simple question to ask: if someone controlled our payment gateway, which of our records would still tell the truth?

The money’s path: correspondents, beneficiaries and screening

Where the money went

The complaint says the fraudulent messages directed the Federal Reserve Bank of New York to transfer funds from Bangladesh Bank’s U.S. dollar account there to specified accounts in the Philippines and Sri Lanka, via specific U.S. correspondent banks. About $81 million reached accounts in the Philippines that had been created in May 2015 in the names of fictitious persons, and the messages carried the fake names with real account numbers. The money was then laundered through multiple bank accounts, a money remitting business and casino junkets, and the complaint says the majority of it had not been recovered. About $20 million sent to Sri Lanka was stopped by the recipient bank and never reached the intended recipient.

The complaint also says that the beneficiary accounts in the Philippines were opened in May 2015, before the theft. The receiving end was ready while the intruders were still waiting inside the sending bank.

What the path can catch

The money’s path is the part of the attack the victim bank does not control. It is controlled by correspondents, intermediaries and receiving banks, each with its own checks. The Sri Lanka transfer shows those checks can work: a recipient bank stopped it. The Philippine transfers show they can fail. For the sending bank, the lesson is about what it can ask of its counterparties and how quickly it can reach them. A bank that discovers a fraudulent message needs named contacts at each correspondent and a tested way to ask for a hold or recall, at any hour.

For banks that act as correspondents or intermediaries, the complaint is a reminder that they are part of other banks’ defenses. A 2020 federal advisory, discussed in the hindsight section, said that this group used unwitting banks, including U.S. banks, as custodians of victim banks’ accounts or as pass-throughs for the fraud. A correspondent that monitors for unusual payment patterns on its respondents’ accounts protects them as well as itself.

Sanctions screening

Treasury’s designation added a separate obligation that runs along the same path. Once the programmer and Chosun Expo Joint Venture were on the list of designated persons, U.S. persons had to block their property and were generally prohibited from dealing with them. For banks, that is a screening question handled by the sanctions compliance function, separate from fraud detection. The two functions look at the same payment flows for different reasons, and the designation was a prompt to make sure each knew what the other was seeing.

Signals the complaint puts on the record

The complaint is also a catalog of signals that a bank’s monitoring could look for. None of them is exotic. Each is something a system recorded or could have recorded at the time.

Failed logins with names that do not exist

The login attempt on the SWIFTLIVE host that used the name of a South American currency exchange is a signal of a kind any gateway can produce: a login attempt with an account name that does not exist on the system. On a host with a short list of legitimate users, every such attempt is worth an alert. The complaint says the intruders deleted some evidence of their login attempts, which argues for sending gateway login records to a system the gateway’s administrators cannot change.

Files named like legitimate tools

The nroff.exe file at the African bank had the same name as a legitimate Alliance Access tool and appeared in the program folder on the day the unauthorized messages were sent. A gateway’s program folders should change only during planned maintenance. Monitoring that records every new or changed file in those folders, and compares it against the vendor’s published files, turns that disguise into a signal.

Database queries for specific transactions

At the African bank, the complaint says, forensic analysis found text files containing database queries to retrieve information about the message with a specific transaction reference number, the same number used in one of the fraudulent messages, created on the day those messages were sent. Other files with the same naming pattern had been overwritten with zeroes. The complaint says the evtdiag.exe malware at Bangladesh Bank had a feature designed to create nearly identical files. Queries against the message database by anything other than the gateway application itself are another signal a bank can monitor.

Preservation by accident

Finally, the evtdiag.exe malware survived, according to the complaint, because bank staff shut down the server on February 5, 2016, so that its self-deletion instruction failed when the server restarted. That was luck. The deliberate version is an incident procedure that preserves the gateway’s memory, disk and logs before anyone restarts or rebuilds it, so that evidence of what was changed is kept. The complaint’s account of how the FBI linked the Bangladesh Bank malware to the group drew on samples like that one.

Designing the independent record

The recommendation at the center of this article, a record the sending endpoint cannot change, can be met in more than one way. Each design has a different cost and a different weakness, and a bank should choose deliberately.

Counterparty data

The strongest independent record comes from outside the bank: statements and confirmations from the correspondent that holds the account, delivered through a channel the gateway does not control. A bank that compares outgoing payments recorded by its own business systems against the correspondent’s account activity can see payments it did not authorize, even if its gateway records have been edited. The weakness is timing: the data arrives after the money has moved, so this check limits loss and concealment rather than preventing the transfer.

An independent copy of message traffic

A second design captures a copy of every message as it leaves the gateway, on a system outside the gateway’s host and administered by different people. The copy can be compared in near real time with what the business systems requested. The weakness is that the capture point must itself be protected; if the intruders reach it, it becomes one more record to edit.

Origination records in the business systems

A third design starts from the other end: every payment must begin as an instruction in a business system, such as a treasury or core banking system, with its own approvals, and every message the gateway sends must match one of those instructions. A message with no matching instruction is, by definition, unauthorized. The weakness is that the matching has to be done somewhere, and if it is done on the gateway, the intruders can defeat it there.

Whichever design a bank chooses, it should test it the way the complaint describes the attack: assume the gateway and its records are under an intruder’s control, and ask whether the independent record would show a payment that nobody in the business requested. A tabletop exercise built on the Bangladesh Bank sequence, from spear-phishing to edited records, is a practical way to run that test with payments, finance, security and the correspondent relationship managers in the room.

In our view, a bank should combine at least two of these designs so that one covers the other’s weakness: matching to origination records before release, and reconciliation against counterparty data after. Neither depends on the gateway telling the truth about itself.

How the four parts cover for each other

The value of separating the four parts is that a bank can see where one part’s failure is caught by another. The complaint gives a test case for each pairing.

  • If the way in fails, as it did, the sending endpoint’s separation and access controls are the next line. At Bangladesh Bank, according to the complaint, the intruders reached the gateway by lateral movement from systems they had compromised through email.
  • If the sending endpoint fails, the bank’s record is the next line, and it works only if it is independent of the endpoint. At all three banks in the complaint’s account, it was not.
  • If the record fails, the money’s path is the last line, and it is in other banks’ hands. The Sri Lanka transfer was stopped there; the Philippine transfers were not.
  • If the money’s path fails, the bank is left with recovery, through law enforcement and its counterparties, and the complaint says the majority of the money sent to the Philippines had not been recovered.

In our reading, the strongest control the victim banks lacked was in the third part: an independent record that would have shown staff, quickly, that messages they had not authorized had been sent. It is also a control entirely in a bank’s own hands.

What the unsealing gave defenders

The Justice Department’s release described the mitigation work that ran alongside the case. It said that throughout the investigation, the FBI and the department gave victims specific information about how they had been targeted or compromised and about the conspiracy’s tactics and techniques, in the United States and abroad, with help from foreign law enforcement partners. It said the FBI had worked with certain private cybersecurity companies to share and analyze information about the intrusion patterns, and that in connection with the unsealing, the FBI and prosecutors gave cybersecurity providers and other private sector partners detailed information on accounts used by the conspiracy, to help their own investigations and disruption efforts.

For a bank, the practical value of a case like this lies in three places. The complaint’s narrative tells a bank what the intruders did, in an order it can test its own defenses against. The indicators shared with providers reach a bank through its security vendors and information sharing groups, if it is connected to them. And the designation puts named persons and an entity on the sanctions list, which reaches the bank through its screening systems.

Each of those channels has an owner inside the bank: the security architecture team for the narrative, the threat intelligence or monitoring team for the indicators and the sanctions team for the designation. September 6, 2018 delivered material to all three at once. A bank that routed the complaint only to its sanctions team, or only to its security team, used a third of it.

Objections

This was a nation-state; no bank can stop that

One objection is that a state-backed team with years of patience will get in anywhere. The complaint supports the first half: the way in was a set of ordinary spear-phishing messages, and they worked. It does not support the conclusion. The intruders needed the sending endpoint and the record to fail together, and both are places where a bank can add controls that do not depend on keeping intruders out of the network.

The messaging network should have caught it

A second objection is that the interbank network should have detected unusual messages. Networks can add checks, and the network’s own security standards are a reasonable thing for a bank to follow, as the 2020 federal advisory recommended. But the messages in the complaint were authenticated messages from the bank’s own gateway, which is what the network is designed to accept. Authorization is the sending bank’s responsibility, and the complaint shows why it has to be checked on the bank’s side.

Independent reconciliation is too slow to matter

A third objection is that reconciliation against a correspondent’s data arrives too late to stop a transfer. Sometimes it does. Even late, it shortens the time between theft and discovery, which we think affects how much can be recalled, and it removes the intruder’s ability to hide the theft from the bank’s own staff. The complaint’s account of the evtdiag.exe malware shows that the intruders planned for concealment to last beyond the transfers themselves.

Our gateway software protects its own records

A fourth objection is that payment software already protects its database against changes, so the attack on the record cannot be repeated. The complaint answers this directly. It says that the Alliance Access file that would ordinarily prevent changes to the message database was modified, in memory at Bangladesh Bank and on disk at the African bank, so that the intruders could change the database. A protection that runs on the same host as the intruder is a protection the intruder can disable. Integrity controls inside the gateway are worth having, and they raise the cost of an attack, but they cannot be the only check on a system the intruder controls.

Attribution changes nothing for the victim

A fifth objection is that charging a North Korean programmer who will not stand trial in the United States does nothing for banks. As a deterrent that may be right. As information it was valuable: the complaint published, in detail, how the intruders worked inside banks, and the FBI said it had shared information with victims and private partners throughout the investigation and, at unsealing, gave cybersecurity providers detailed information on accounts the conspiracy used.

How it applies across institutions

Large banks and small ones

A large bank may run its own gateway with a dedicated team. A smaller bank may rely on a service bureau or a third party for its connection, as the Vietnamese bank in the complaint did. The four parts apply to both, but ownership shifts. For a bank that uses a third party, the record its staff read may be produced on the third party’s systems, and the independence question has to be asked of the provider: who can change what our staff see, and what record would we have if the provider’s environment were compromised?

Banks that rely on a provider

The Vietnamese bank in the complaint shows how a third-party arrangement changes the record question. Its staff never saw the gateway. They saw documents on the provider’s server, opened with software the intruders had replaced. For a bank in that position, the independence of the record cannot be designed inside the bank alone. The bank has to know, from the provider, where its message records are produced and stored, who can change them, and how the provider would detect tampering. It also needs a record of its own that does not pass through the provider, and counterparty data is the natural candidate.

Contracts with such providers deserve review for this purpose. They should say how quickly the provider will tell the bank about a compromise of the environment that serves it, whether the bank can obtain an independent copy of its own message traffic, and how evidence will be preserved for the bank and for law enforcement if something goes wrong.

Other payment systems

The complaint concerns interbank messaging, but the structure applies to any system where a compromised endpoint can both send instructions and shape the record: domestic wire systems, ACH origination platforms, card payment switches and treasury management systems. The FASTCash alerts that followed, discussed below, show North Korean government hackers turning to the servers that authorize card withdrawals.

Outside banking

Health systems and public agencies make large payments too, to vendors, contractors and benefit recipients, and their payment platforms can be compromised in the same way. For them, the third part transfers directly: a payment record that the payment system cannot change, reconciled against the bank’s statement by people who do not depend on the payment system.

What we would ask a payments leader today

  • If an intruder controlled our payment gateway for a day, which of our records would still show the truth, and who would read them?
  • Is the gateway separated from the general network, with its own strong authentication, monitored login records and a short list of accounts that can create and release messages?
  • Do we reconcile what we sent against data from our correspondents or counterparties that the gateway cannot change, and how soon after sending?
  • Do we have named contacts and a tested process to ask each correspondent to hold or recall a payment, at any hour?
  • If a third party runs our connection, have we asked it the same questions?

With hindsight

North Korean government hackers were soon visible in a different payment system. On October 2, 2018, DHS, Treasury and the FBI issued a joint technical alert, TA18-275A, on what the government called the FASTCash campaign: a North Korean scheme to cash out money from automated teller machines. The alert listed retail payment systems as the systems affected and provided indicators of compromise so institutions could detect the malware.

On August 26, 2020, CISA, Treasury, the FBI and U.S. Cyber Command issued a joint advisory, AA20-239A, titled FASTCash 2.0: North Korea’s BeagleBoyz Robbing Banks. It described a North Korean team dedicated to robbing banks through remote internet access, which it said overlapped to varying degrees with groups tracked by industry as Lazarus, APT38, Bluenoroff and Stardust Chollima. It said the team was responsible for the FASTCash cash-outs reported in October 2018 and for fraudulent abuse of compromised bank-operated SWIFT system endpoints since at least 2015. It said that, according to public estimates, the team had attempted to steal nearly $2 billion since at least 2015. It also said that in 2018 a bank in Africa could not resume normal ATM or point of sale services for almost two months after an attempted FASTCash incident, and that in 2018 the team deployed wiper malware against a bank in Chile that crashed thousands of computers and servers to distract from efforts to send fraudulent messages from the bank’s compromised SWIFT terminal.

The advisory said the team appeared to seek two specific systems once inside a financial institution: the SWIFT terminal and the server hosting the payment switch application. It said the team used unwitting banks, including banks in the United States, as custodians of victim banks’ accounts or as pass-throughs for the fraud. Its recommendations for all financial institutions included verifying compliance with the FFIEC handbooks and with industry security standards, including SWIFT’s customer security controls framework. For institutions with retail payment systems, it recommended multi-factor authentication for anyone accessing the switch application server, isolation of payment infrastructure and a baseline of expected transaction participants, amounts, frequency and timing, with anomalous transactions flagged.

On February 17, 2021, the Justice Department unsealed an indictment charging the programmer named in the 2018 complaint and two other North Korean programmers, whom it described as members of units of the Reconnaissance General Bureau, a North Korean military intelligence agency. It said the units were known in the cybersecurity community by names including the Lazarus Group and APT38, and that one of the three had previously been charged in the complaint unsealed in September 2018. The indictment alleged attempts from 2015 through 2019 to steal more than $1.2 billion from banks in Vietnam, Bangladesh, Taiwan, Mexico, Malta and Africa by hacking the banks’ networks and sending fraudulent SWIFT messages, and ATM cash-out thefts including the October 2018 theft of $6.1 million from BankIslami Pakistan Limited. The department described the overall conspiracy as aiming to steal and extort more than $1.3 billion of money and cryptocurrency.

What held up from the 2018 complaint is the picture of how the theft worked: intruders who reached the payment endpoint and then shaped the bank’s own record of what it sent. The later advisory’s account of a wiper used to distract staff in Chile fits the same pattern of attacking the bank’s view of events. What changed is the range of targets, from interbank messaging to card switches and cryptocurrency. We think the defense that applies across all of them is the one the complaint pointed to: records and checks that the compromised system cannot alter.

Spotted an error? Tell us and we will check it and correct the article where needed.

  • Bangladesh Bank
  • SWIFT
  • Lazarus Group
  • payment fraud
  • reconciliation
  • OFAC