Nine CryptoPro Vulnerabilities Turn ATM Security Into a Software Supply Chain Problem

A security researcher who has spent five years probing automated teller machines just showed how a single piece of encryption software can turn a narrow hardware research project into a much bigger story about how patches actually move through the software supply chain.

At Black Hat USA and DEF CON in Las Vegas earlier this month, Matt Burch, a principal security researcher at Atredis Partners, presented nine vulnerabilities he found in CryptWare’s CryptoPro Secure Disk, a full-disk encryption and pre-boot authentication product for Windows. CryptoPro is sold to ATM manufacturers, including Diebold Nixdorf, a major ATM manufacturer, and it is also licensed far outside banking, to other embedded-device makers and to large organizations running Windows fleets. That dual audience is what turned Burch’s ATM research into a supply chain story, first reported by Dark Reading in July, corroborated by SC Media, and expanded on today by WIRED.

How the vulnerabilities work

Burch’s findings target the moment before an ATM’s operating system fully loads. In certain fail states, CryptoPro’s pre-boot decryption process would default to mounting the encrypted volume in plaintext instead of refusing to load. Combined with what Burch described as a superficial check for whether a disk was using LUKS encryption, an attacker could trick the software into treating an unencrypted disk as though it were properly secured. CryptoPro also stored key material and configuration values on the same disk it was supposed to protect, and its Secure Boot implementation could be coaxed into running an attacker’s own Linux environment instead of the vendor’s intended one.

Chained together, those bugs let Burch run his own code during pre-boot, recover the keys needed to decrypt the machine’s Windows environment, and then steal cash using a standard jackpotting technique, the class of attack that forces an ATM to dispense its stored cash on demand.

Why the top half of an ATM is the soft target

Burch’s research also underscores a physical design tradeoff most ATM operators take for granted. The safe holding the cash is heavily armored, but the head unit above it, which houses the PC hardware, is not. “[Manufacturers] don’t necessarily consider the top portion of the ATM where all the PC components are a significant security risk,” Burch told Dark Reading. “It’s constructed with lower-grade steel,” or sometimes plastic. The door is typically held by a cable-and-solenoid mechanism that can be pried or damaged open, giving an attacker physical access to the computer inside.

Once inside, Burch explained, the dispensing process depends on a single dynamic link library he refers to as XFS, an extension for financial services that connects the machine to the bank’s network and authorizes a payout. Once past CryptoPro’s pre-boot encryption layer, an attacker is inside the same Windows environment that ultimately talks to XFS, the chain Burch says let him run his own code, recover decryption keys, and dispense cash through jackpotting. The best-known jackpotting malware family is called Ploutus. Jackpotting attacks broadly have been reported to the FBI roughly 2,000 times since 2020, Dark Reading reported, with more than 700 cases in 2025 alone accounting for over $20 million stolen.

A disagreement over how much of this reaches real ATMs

The vulnerabilities themselves are not in dispute. Whether they matter to a live Diebold Nixdorf ATM is. Diebold Nixdorf’s hard disk encryption component, part of its Vynamic Security Suite, incorporates elements of CryptoPro, and CryptoPro Secure Disk has been listed in Diebold’s own end user license agreements since at least 2018. Burch has called CryptoPro “foundational” to that security suite.

Diebold Nixdorf disagrees, at least in degree. Spokesperson Michael Jacobsen told WIRED that only two of Burch’s nine vulnerabilities are relevant to Diebold’s hard disk encryption implementation, and that the company shipped fixes for those two in December 2025, though he maintains they could not have been exploited on their own to compromise a Diebold Nixdorf ATM. In an earlier statement to Dark Reading, Jacobsen was more categorical, saying the findings pose “little to no additional risk” in a real-world environment, while still conceding that two of the nine vulnerabilities are “theoretically applicable to our HDE under certain conditions.” Jacobsen declined to clarify exactly how Diebold implements CryptoPro when asked. Burch speculates the company may be relying on one of CryptoPro’s alternative cryptographic mechanisms rather than the configuration he tested.

The real bottleneck is getting a fix onto every machine

Whichever framing is right, CryptoPro’s reach outside ATMs is the bigger structural problem. According to Dark Reading, the vendor claims more than 500,000 licenses in use across five continents and 20 industries. CryptWare, a German company, patched all nine bugs in two phases: version 7.7.2 in November 2025 and version 7.7.3 in December 2025. Burch says CryptWare was prompt and collaborative throughout disclosure, and he personally validated that the patches closed the holes he found. CryptoPro does not appear to publish public release notes, so Burch believes CryptWare reached out to customers directly to share patch information.

That gap between a fix existing and a fix reaching every affected machine is the software supply chain problem in practice. A vulnerability gets patched at the source, but every company that builds the flawed component into its own product then has to build, test, and ship its own tailored update, and every one of that company’s customers has to actually install it. Jacobsen described Diebold Nixdorf’s version of that process to WIRED: assess impact, identify affected products and configurations, build updates through its product security and engineering teams, then notify customers through channels such as its Global Security Portal, with deployment on live ATMs coordinated around each customer’s own service agreements and change-management process. For a fleet of machines that are expensive to visit and cannot always be paused for maintenance, that last step, someone actually applying the patch, is often the slowest part of the chain.

Obscurity is a weaker defense than it used to be

Burch’s closing argument is less about ATMs specifically and more about why niche, unglamorous security products like CryptoPro tend to go unexamined for years. Security through obscurity, hiding a product from scrutiny rather than hardening it, has always been a weak strategy, and Burch argues AI tools are eroding whatever protection it still offered. “AI really blows away the obscurity model,” he told WIRED. “You don’t need to fully understand how something works anymore to move forward and potentially have a big impact.” His broader point, he said, is a simple one: “It’s important to ensure that whenever you’re using cryptography or encryption, the secrets are also secured and not easily generated or recovered. If you give someone a lockbox and then the key is right next to the box, it defeats the purpose of having the box locked to begin with.”

The same lesson extends well past banking. CryptoPro’s own marketing claims license holders in 20 industries, and any organization running a third-party product built on it, embedded-device makers included, inherits the same patching lag between a vendor’s fix and its own customers actually installing it.