Contagious Interview (tracked as WaterPlum) is a North Korean actor that poses as an employer and gets developers to run malicious code as part of a coding assignment. That’s according to the joint advisory of 18 September 2026 from the FBI, DC3, ASD’s ACSC, BND, BfV, Japan’s NPA and Japan’s National Cybersecurity Office (NCO).
TL;DR:
- At least 30,000 PCs in over 100 countries were exploited from about December 2025 through July 2026 (per the advisory)
- The advisory’s VS Code advice: answer “No” to the trust prompt, which keeps the folder in Restricted Mode and prevents execution of
.vscode/tasks.jsonon launch - Read
.vscode/tasks.jsonand thepackage.jsonscripts before you run anything; if you can’t explain a command, don’t run it - Run unknown code only inside a sandbox or virtual machine, per the advisory
What the advisory says about fake interviews and coding tests
The actors impersonate legitimate AI, cryptocurrency, or NFT companies and have also used recruiting services. They recruit through social media, online job platforms, gig work platforms, or freelance marketplaces (per the advisory). If you’re a Web3 or blockchain developer on a freelance board, the inbound message can look exactly like a genuine offer.
The ask is concrete: download and execute files hosted on code-repository or collaboration platforms to “complete a coding assignment” or “troubleshoot an error” in the video call. The files are malicious Node Package Manager (NPM) packages carrying BeaverTail, InvisibleFerret, OtterCookie, OtterCandy, or StoatWaffle malware and related variants. Nobody asks you to install an obscure driver. They ask you to run a repo.
From around December 2025 through July 2026, WaterPlum exploited at least 30,000 PCs in over 100 countries, including Japan and the United States. The advisory names Europe as a victim region. The actors exfiltrated funds or credentials from over 7,000 cryptocurrency wallets, taking at least 1.7 billion JPY (10.71 million USD) on behalf of DPRK.
The malware reaches for authentication data stored in your browser, clipboard contents, key-logs, screenshots, private keys and seed phrases, and ID documents like driver’s licenses or passports kept on a PC or in shared folders. The advisory adds that stolen credentials can be used to exfiltrate crypto assets, personal data, or trade secrets from your employer, clients, or contracting parties.
So the step to stop at is “run the repo”. If the assignment needs you to execute code, the advisory’s own mitigation is a sandbox or virtual machine, not your daily machine.
How a project folder can run code before you read it
You clone a repo from an interviewer and open it in VS Code. On current defaults, the folder opens in Restricted Mode with a trust banner, so it’s already restricted before you’ve read a line. The risky moment is the click that follows: Trust in the banner, or Trust in the Manage menu. That click is the checkpoint between “folder on disk” and “folder executing”. Some setups still show an upfront dialog; if you get one, answer No (my advice, following the advisory’s wording).
The mechanism is in the advisory’s footnote on StoatWaffle. This malware family uses blockchain-themed project repositories as decoys, and the footnote describes a malicious VS Code configuration file that triggers auto-run code execution once the folder is opened and trusted (per the advisory). The folder itself is the trigger.
The VS Code docs make the task angle equally concrete. Task definitions live in the committed .vscode folder, so they’re part of the repo and shared with everyone who clones it. The docs put it plainly: if someone creates a malicious task, anyone who cloned that repository could run it unknowingly (per VS Code docs, Workspace Trust).
Open it in Restricted Mode and stay there while you read
The advisory tells you to answer “No” to the “Do you trust the author of the files in this folder?” prompt. That keeps you in Restricted Mode and prevents execution of .vscode/tasks.json on launch.
What Restricted Mode limits, per the VS Code docs:
- Terminal and workspace settings are disabled or limited.
- AI agents are disabled or limited.
- Tasks: running or listing them triggers a trust-confirmation prompt.
- Debugging is disabled.
- Extensions that haven’t opted into Workspace Trust are disabled by default.
A banner at the top links to the Workspace Trust editor. Staying in Restricted Mode neutralises the StoatWaffle mechanism: untrusted folder, no trigger.
Parent-folder trap. Trust applies to every subfolder of a trusted parent (per VS Code docs). If ~/projects is trusted and you clone a take-home into ~/projects/acme-assignment, it opens trusted. The advisory warns against opening unknown projects from a path you previously marked trusted. Clone take-homes somewhere outside any trusted parent, like ~/takehomes if you’ve never trusted it.
What it doesn’t cover. A malicious extension you installed yourself can execute code and ignore Restricted Mode (per VS Code docs). Restricted Mode is a VS Code feature, not an OS boundary, so it doesn’t replace a sandbox or VM.
Don’t pass --disable-workspace-trust on the command line. It switches the feature off for the session, and you have no reason to do that while reading an unknown repo.
What to read before you run npm install
The advisory says to verify the contents of .vscode/tasks.json in Restricted Mode or with another editor, looking for code that downloads or executes additional files, and to check for obfuscated or unreadable sections before execution. Here’s the order I’d follow:
-
Read
.vscode/tasks.json, every command line. If a task pipescurloutput into a shell or decodes abase64blob, stop. The advisory flagscurl,base64,-enc,mshta, andhiddenas red-flag strings, and says to avoid running scripts unless you fully understand them. You don’t need to reverse-engineer the payload. Recognise the download-and-execute pattern and walk away. -
Read
package.json, especially thescriptsblock and the dependency list. My advice, not a sourced fact: npm packages can run scripts at install time, so read those script names and the dependency names before you typenpm install. A script that downloads and executes something has the same shape as the malicious NPM packages the advisory describes. If the script body is hard to read, stop. -
Scan the rest of the repo for obfuscated sections. A block of unreadable hex in a
.jsfile isn’t a bug. It’s a reason to stop. If you can’t tell what a section does, you can’t clear it. -
Check where the repo came from. My view: a repo on a brand-new account with a single project carries a different risk than one in a company’s public organisation. Given the recruitment channels above, the source matters.
-
If you can’t explain a command in plain language, don’t run it. No “I’ll just try it and see.” That holds whether the command sits in a task, a script, or a one-liner pasted into the interview chat.
Run it in a VM or a disposable container, not on your work laptop
The advisory’s sandbox-or-VM line comes with a second rule: avoid executing code from untrusted third parties on machines handling cryptocurrency assets or personal data. That covers your work laptop, the machine with a wallet browser extension, and the box where your SSH keys sit in ~/.ssh.
My recommendation, not a sourced requirement: if you have crypto wallets, a company-issued machine, or SSH keys in your home directory, spin up a throwaway VM, snapshot it before you start, and delete the image when the assignment is done. A disposable container is acceptable only if you mount nothing sensitive (no home directory, no SSH agent socket, no cloud credential files, no Docker socket) and you drop it after use.
Also my advice: don’t run the repo on a machine or account where your logged-in browser sessions live. Browser-stored authentication data is on the advisory’s theft list above, so the code you run and the profile holding your session tokens should never share a machine.
How the three options compare (my judgement, not a sourced spec):
| Option | What it protects | Catch |
|---|---|---|
| VS Code Restricted Mode only | Blocks automatic execution of tasks, debugging, and extensions that haven’t opted into Workspace Trust (per VS Code docs) | A setting on the same machine, not a separate environment; doesn’t stop an extension you installed or anything you run in an external terminal |
| Disposable container | Depends entirely on what you mount into it | A mounted home directory or Docker socket undoes the separation |
| Throwaway VM | A separate machine image you can delete after the session | Setup time: provisioning, snapshot, teardown |
Red flags in the interview itself
Watch the “troubleshoot an error” framing from the advisory. A hypothetical example: “The call is glitching, run this script to fix the audio.” It’s the same vector as “here’s your take-home, clone this branch”. If someone on a video call asks you to execute something to fix the call, hang up.
The face-swap detail makes the call itself unreliable. Per the advisory, WaterPlum actors conducted interviews using AI face-swapping software, then after a few minutes disabled their video and told the target to do the same, citing network issues. Now both cameras are off, the person is asking you to run code, and the identity check is gone.
Binance’s own incident report shows how it plays out. A software engineer was contacted through a legitimate job-hunting platform by an individual posing as a project director. The second interview, with someone claiming to be the CEO of a fictitious company, ran on a popular social chat app. Near the end, the engineer was instructed to download an “assignment” containing a malicious code repository onto his work computer. That code executed during the setup process (per Binance, 3 November 2026). A company machine took the hit, not a personal laptop. Binance’s report states that the latest information released by the FBI indicates North Korea is responsible for the incident (per Binance, 3 November 2026). Binance does not tie the case to the actor in this advisory, so treat them as separate until a source says otherwise.
My advice before you run anything from a repo that arrived mid-interview: ask for the company’s registered email domain, check it yourself on WHOIS or the company’s own site, and send a short confirmation from a channel you opened. If the “interviewer” can’t give you a domain you can verify, slow down. Separately, how much free work an employer may ask for is a different question, covered in unpaid take-home assignments.
If you already ran it
Disconnect the device from the network first. The advisory is direct: if you suspect compromise, immediately disconnect the impacted device from the internet to disable external communications (per the advisory). The StoatWaffle family in the same advisory includes a RAT for persistence and pivoting, so a connected machine may still be communicating.
Assume the data is already gone. The advisory says to assume data, including cryptocurrency wallet information, may already have been exfiltrated, even after detection and removal. Create a new wallet on a separate device, transfer all assets to it, and store the new seed phrase offline. If the old seed lives anywhere on the infected machine, treat it as compromised.
Back up essential data, then fully reset the operating system. The advisory warns that undetected malware may remain, so a clean scan from the same OS isn’t a fresh install.
My recommendation beyond the advisory: change your passwords and revoke active sessions, from a different, clean device.
Did that machine have access to your employer’s systems? Then tell their security team the same day. That step is my addition, not the advisory’s. The breach is downstream of your laptop, and they’d rather hear it from you than from an alert.
FAQ
Does VS Code Restricted Mode stop a malicious .vscode/tasks.json?
The joint advisory says Restricted Mode prevents execution of .vscode/tasks.json on launch, and the VS Code docs say running or even listing tasks in Restricted Mode asks you to confirm trust first (per the advisory, 18 September 2026, and VS Code docs, Workspace Trust). It is not a full sandbox. For code you plan to run, use a VM or a disposable container.
How do I open a take-home repo in Restricted Mode?
On current defaults a new folder already opens in Restricted Mode with a trust banner, so the risky click is Trust. Don’t click it, and answer No if a prompt asks whether you trust the authors (the docs name the option “No, I don’t trust the authors”). Clone the repo outside any parent folder you already trust, since trusting a parent trusts every subfolder under it.
Is it safe to run npm install on a take-home assignment?
Only after you’ve read what it will run. Open .vscode/tasks.json and the scripts block in package.json, look for download-and-execute patterns, and stop if you can’t explain a command. For anything you do run, the advisory says to use a sandbox or virtual machine.
Is a container enough, or do I need a VM?
The advisory only says “sandbox or virtual machine”. My view, not a sourced fact: a throwaway VM is the better route if your laptop holds wallets, work credentials or SSH keys. A container is acceptable only if you mount nothing sensitive and delete it afterwards.
What should I do if I already ran a suspicious assignment?
Disconnect the device from the internet, assume data including wallet information may already be exfiltrated, move any assets to a new wallet created on a separate device and store the new seed phrase offline, then back up what you need and fully reset the operating system (per the advisory, 18 September 2026). If the machine had work access, also tell your security team the same day (my addition, not the advisory’s).
Related articles
- The take-home assignment playbook
- Unpaid take-home assignments under EU labor law in 2026
- Spotting ghost job postings in 2026
Sources
All pages last checked on 15 November 2026.
- North Korean “WaterPlum,” commonly referred to as “Contagious Interview,” Cyber Actor Group Targeting IT Professionals — FBI IC3 with DC3, ASD’s ACSC, BND, BfV, Japan NPA and Japan NCO, 18 September 2026 — at least 30,000 PCs in over 100 countries from about December 2025 through July 2026; over 7,000 wallets; at least 1.7 billion JPY (10.71 million USD); the Restricted Mode and
.vscode/tasks.jsonadvice; “Only run unknown code inside a sandbox or virtual machine”; StoatWaffle footnote on malicious VS Code projects; the fake-interview behaviours quoted above - Workspace Trust — Visual Studio Code docs — Restricted Mode limits AI agents, terminal, tasks, debugging, workspace settings and extensions; tasks live in the committed
.vscodefolder; parent-folder trust;--disable-workspace-trustaffects the current session only; a malicious extension can still run code and ignore Restricted Mode - 29 October 2026 Binance Security Incident: Initial Findings and Response Update — Binance, 3 November 2026 — September 2026 contact through a legitimate job-hunting platform by an individual posing as a Project Director; second interview with an individual claiming to be the CEO, on a popular social chat app; “assignment” containing a malicious code repository downloaded onto his work computer; code executed during the setup process
Try Xeito
Juggling several interview processes at once? Try Xeito: the kanban tracker keeps notes on each application, so you can log who sent which take-home and where you ran it, and the browser extension saves a listing from the job board with an EU-workability badge that sits next to Apply on the listing page. Xeito doesn’t scan repositories and isn’t a security tool.
This article is for informational purposes only and does not constitute legal or tax advice. Security guidance changes; confirm the current recommendations in the joint advisory and the VS Code documentation before you rely on them.