Security
Four Governments Trace 30,000 Infected Developer PCs to Fake Job Interviews
Police and intelligence agencies in Japan, the US, Australia and Germany say North Korea's WaterPlum group took ¥1.7 billion from more than 7,000 crypto wallets in eight months. Its newest tool needs one click from the victim: telling VS Code the folder can be trusted.
The job offer comes from what looks like an AI, crypto or NFT company. There is a video interview, then a coding test. The candidate is told to clone a project and run it, or to download a fix because the video call is misbehaving. By the time the interview ends, the laptop belongs to someone else.
That is the pattern described in a joint advisory published on September 18 by seven agencies in four countries: Japan's National Police Agency and National Cybersecurity Office, the FBI and the Defense Department's Cyber Crime Center, Australia's ACSC, and Germany's foreign intelligence service (BND) and domestic security agency (BfV). The group is the one widely known as "Contagious Interview." The advisory calls it WaterPlum.
The numbers attached to it are new. According to the National Police Agency's figures in the advisory, between around December 2025 and July 2026 WaterPlum compromised at least 30,000 PCs in more than 100 countries, including Japan and the United States. It took funds or account credentials from over 7,000 cryptocurrency wallets and moved at least ¥1.7 billion, which the agencies put at $10.71 million, to North Korea. The primary targets, the agencies write, were individual web designers, engineers and specialists in crypto, blockchain and Web3.
Who they say is behind it
The attribution is stated plainly. The NPA and the FBI assess that WaterPlum's operators, and some of North Korea's overseas IT workers, work under the 313 General Bureau of the Munitions Industry Department, subordinate to the Central Committee of the Workers' Party of Korea. The advisory ties the two populations together with one concrete piece of evidence: WaterPlum actors and North Korean IT workers used the same IP addresses to reach laptop farms, to use crowdsourcing services, and to apply for jobs at a Japanese crypto exchange.
That exchange episode is the advisory's only named case study, though the exchange itself is not named. In May 2025 it received an application for an engineering role, submitted through a VPN, with a résumé listing ten or more skills in each category and a quick run of jobs across Europe and Asia. In the interview the applicant could not discuss most of it. The company did not hire him, and the advisory says no damage occurred. The acknowledgements section thanks two private partners, NTT Security Japan and the exchange bitFlyer; the advisory does not say whether bitFlyer was the exchange in the case study, and coverage that names it as such goes beyond the text.
Japanese authorities also say that, for the first time in Japan, they identified and dismantled a "laptop farm" run by a local enabler: a home where employer-issued computers are kept switched on and operated remotely by workers in North Korea, China or Russia. They found evidence that the group behind it sent several hundred million yen in crypto abroad. The advisory does not say whether anyone has been charged.
The five tools, and the one that needs a single click
The malware list will be familiar to anyone who follows this group. BeaverTail is JavaScript hidden in npm packages. InvisibleFerret is a Python backdoor. OtterCookie is a JavaScript remote-access trojan and stealer, and OtterCandy combines it with another RAT. They collect browser-saved passwords, clipboard contents, keystrokes, screenshots, ID photos and wallet private keys and seed phrases.
The fifth, StoatWaffle, is the one worth slowing down on. The advisory describes it as a modular Node.js family delivered through malicious Visual Studio Code projects, using blockchain-themed repositories as bait. The project carries a VS Code configuration file that, in the advisory's words, triggers auto-run code execution "when the folder is opened and trusted by the victim." The mitigation section names the file: .vscode/tasks.json.
That makes the trust prompt the whole defence. VS Code asks whether you trust the authors of a newly opened folder. Answer no and it opens in Restricted Mode, which, per Microsoft's documentation, disables tasks, the terminal, debugging, workspace settings, extensions and AI agents. Answer yes and the project's tasks can run. The same documentation notes that trusting a parent folder extends trust to every subfolder, which explains one of the advisory's less obvious warnings: do not open an unknown project from any location you have previously marked as trusted. A developer who once trusted ~/code will never see the prompt for a test repo cloned into it.
WaterPlum, by the advisory's own figures
| Measure | Figure |
|---|---|
| Period | Around Dec 2025 – Jul 2026 |
| PCs compromised | At least 30,000 |
| Countries | More than 100 |
| Wallets with funds or credentials taken | Over 7,000 |
| Crypto moved to North Korea | At least ¥1.7 billion ($10.71 million) |
| Malware families named | 5 |
What the numbers do and do not say
Two cautions on the figures. First, they count different things. Thirty thousand is machines; seven thousand is wallets, and one machine can hold several. The advisory gives no count of victims whose wallets were emptied, so the two numbers cannot be divided into a success rate. Second, $10.71 million is small by North Korean standards, and that is the point rather than a flaw. Spread over more than 7,000 wallets, it averages under $1,530 each. This is volume work against individuals, not a single exchange heist, and the advisory is explicit that the money is only part of the take. Stolen ID photos can let IT workers impersonate victims, and stolen credentials can open the victim's employer or clients.
The advisory also includes some unusually human details. WaterPlum members ran interviews through AI face-swapping software, then after a few minutes switched off their video and told the candidate to do the same, blaming the network. They practised Japanese pronunciation with text-to-speech. They used free tiers of translation and AI services. On North Korean public holidays they played games and watched football instead of working.
What the agencies tell developers to do
The advice is specific enough to act on. Run code from strangers only in a sandbox or virtual machine, never on a machine that holds crypto or personal data. Treat scripts containing strings like curl, base64, -enc or mshta as suspect until understood. Open unknown VS Code projects in Restricted Mode and read tasks.json before trusting anything. If a machine is infected, assume the wallet is already gone: create a new wallet on a separate device, move everything to it, and reinstall the operating system.
The Take
Seven agencies signing one document is a statement, and the laptop-farm takedown in Japan is real enforcement. But the most useful line in nine pages is a note about a dialog box. Crypto developers have spent years being told that the danger is a malicious dependency buried five levels deep. WaterPlum's newest tool does not need that. It needs a candidate who wants the job, a repository that looks like a normal take-home test, and one click on "Yes, I trust the authors." Editors that run a project's own configuration on open are a convenience that ends up working as an attack surface, and the same idea keeps turning up in coding agents. Until the tools default the other way, the trust prompt is the security boundary, and nobody treats it like one. If a recruiter's first technical request is to run their code on your machine, that tells you what the interview is for.