Threat reportSupply ChainTL-2026-2150
24 Malicious npm Packages Abuse Registry Mirrors as Phishing Infrastructure (Fake Cloudflare/Microsoft Login Pages)
OX Security identified 24 npm packages (published 2026-08-04 to 2026-08-24) whose only payload is a single HTML file impersonating a Cloudflare Turnstile verification screen; obfuscated JavaScript in the page queries a redirect target and sends visitors on to phishing or attacker-chosen infrastructure.
Key facts
| Identifiers | TL-2026-2150 |
|---|---|
| Severity | MEDIUM |
| First disclosed | |
| Affected products | npm, Inc. / OpenJS Foundation npm public registry and its downstream CDN mirrors (unpkg, npmmirror, Tencent Cloud npm mirror, Yarn registry mirror) |
| Weaknesses (CWE) | CWE-506, CWE-451 |
| Tools | Cloudflare Turnstile (spoofed UI) |
| MITRE ATT&CK techniques | T1027 Obfuscated Files or Information, T1036 Masquerading, T1195 Supply Chain Compromise, T1204 User Execution, T1566 Phishing, T1583 Acquire Infrastructure, T1608 Stage Capabilities |
| Motivation | FINANCIAL |
| Target sectors | technology, cross-sector |
| Target regions | Global |
| Detection rules | 9 (1 shown free) |
| Sources | 7 cited sources |
Detection coverage
As of 2026-08-26, Threadlinqs Intelligence publishes 9 detection rule(s) for TL-2026-2150 across Splunk SPL, Microsoft KQL and Sigma, covering 20 indicator(s) of compromise. The whole corpus is readable without an account. Every threat comes with one complete detection in Splunk SPL, Microsoft KQL and Sigma, free for everyone; a free account adds daily TL Creds to unlock the rest. Paid plans add a larger daily allowance, correlation and the MCP server. Threadlinqs MCP server · View plans.
Free detection: Direct Browser Retrieval of Raw index.html from npm/unpkg Mirror CDN (Fake Turnstile/MS Login Delivery)
Detects direct HTTP GET requests to npm registry mirror CDNs (unpkg.com, npmmirror.com, Tencent Cloud npm mirror, Yarn registry mirror) for a package's raw index.html file — the delivery mechanism OX Security documented for 24 npm packages whose only payload is a fake Cloudflare Turnstile or Microsoft login page. Legitimate package-manager traffic requests tarballs/metadata via registry.npmjs.org or resolves through npm/yarn/pnpm clients with characteristic Accept/User-Agent headers; a browser navigating straight to a mirror's /index.html path is the phishing-page-render trigger.
- Languages: Splunk SPL, Microsoft KQL, Sigma
- ATT&CK: T1608, T1195, T1566, T1036
- Data sources: web_proxy, zscaler, palo_alto_threat, dns
- Severity: high
- Confidence: high
Splunk SPL
index=proxy (sourcetype=squid:access OR sourcetype=zscaler:web OR sourcetype=pan:threat) dest_host IN ("unpkg.com", "npmmirror.com", "registry.npmmirror.com", "mirrors.cloud.tencent.com", "registry.yarnpkg.com")
| rex field=uri_path "(?<pkg_path>[^/]+@[0-9\.]+/index\.html)"
| where isnotnull(pkg_path)
| eval is_browser_ua=if(match(http_user_agent, "(?i)(Mozilla|Chrome|Safari|Edg)") AND NOT match(http_user_agent, "(?i)(npm|yarn|pnpm|node-fetch|axios)"), 1, 0)
| where is_browser_ua=1 AND method="GET" AND (status=200 OR status=304)
| stats count min(_time) as first_seen max(_time) as last_seen values(uri_path) as paths by src_ip, dest_host, http_user_agent
| where count>=1
| sort -last_seenMicrosoft KQL
DeviceNetworkEvents
| where Timestamp > ago(24h)
| where RemoteUrl has_any ("unpkg.com", "npmmirror.com", "registry.npmmirror.com", "mirrors.cloud.tencent.com")
| where RemoteUrl matches regex @"[^/]+@[0-9.]+/index\.html"
| where InitiatingProcessFileName in~ ("chrome.exe", "msedge.exe", "firefox.exe", "iexplore.exe", "brave.exe")
| project Timestamp, DeviceName, InitiatingProcessAccountName, InitiatingProcessFileName, RemoteUrl, RemoteIP
| summarize FirstSeen=min(Timestamp), LastSeen=max(Timestamp), Hits=count(), Paths=make_set(RemoteUrl) by DeviceName, InitiatingProcessAccountName, InitiatingProcessFileName
| order by LastSeen descSigma
title: Direct Browser Access to npm Mirror CDN Raw index.html (Fake CAPTCHA/Login Phishing Delivery)
id: 9c3f2a4e-1b6d-4e7a-8f21-6d0b5a9c2e11
status: experimental
description: Detects a browser process fetching a package's raw index.html directly from an npm mirror CDN (unpkg, npmmirror, Tencent Cloud npm mirror), matching the OX Security-documented technique of hosting fake Cloudflare Turnstile / Microsoft login phishing pages as npm package payloads.
references:
- https://www.ox.security/blog/research-clickfix-phishing-npm-packages/
- https://thehackernews.com/2026/08/24-npm-packages-abuse-unpkg-mirrors-to.html
author: Threadlinqs Team
date: 2026-08-26
tags:
- attack.resource-development
- attack.t1608
- attack.initial-access
- attack.t1195
- attack.t1566
- attack.defense-evasion
- attack.t1036
- cwe.506
- cwe.451
logsource:
category: proxy
product: web
detection:
selection_host:
c-uri|contains:
- 'unpkg.com'
- 'npmmirror.com'
- 'registry.npmmirror.com'
- 'mirrors.cloud.tencent.com'
selection_path:
c-uri|re: '[^/]+@[0-9.]+/index\.html'
filter_client:
cs-user-agent|contains:
- 'npm/'
- 'yarn/'
- 'pnpm/'
- 'node-fetch'
- 'axios'
condition: selection_host and selection_path and not filter_client
falsepositives:
- Developers or security researchers intentionally auditing a package's raw mirror contents in a browser
- Automated SBOM/dependency-scanning tools that render or fetch HTML payloads for static analysis
level: mediumKnown false positives:
- Developers or researchers manually browsing to a package's mirror URL to audit its contents
- Automated dependency/SBOM scanners that fetch and render package HTML assets for static analysis
Tuning notes: Extend the dest_host/RemoteUrl allowlist as new mirror hosts are confirmed (e.g. additional Yarn/Verdaccio mirrors); allowlist known internal SBOM/SCA tooling user-agents to cut noise; narrow to first-seen-per-host in the last 24h once a baseline of legitimate mirror traffic is established.
- Sentinel: Browser Navigation to Spoofed Cloudflare Turnstile / Microsoft Login Page Hosted on npm Mirror · unlock with TL Creds
- Sentinel: Clipboard-Paste Command Execution Chained to Fake Turnstile/Login Page Visit · unlock with TL Creds
- Sentinel: keyval.org Dead-Drop Query Chained to Newly-Resolved External Redirect Domain · unlock with TL Creds
- Sigma: Web Proxy Access to Malicious npm-Mirror-Hosted Phishing HTML · unlock with TL Creds
- Sigma: ClickFix Encoded-Command Execution via Explorer-Spawned Interpreter · unlock with TL Creds
- Sigma: DNS/Proxy Query for keyval.org Dead-Drop Resolver Endpoint · unlock with TL Creds
- ClickFix-Style Run-Dialog Command Execution Following npm-Mirror Phishing Page Visit · unlock with TL Creds
- Dead-Drop Resolver Callback to [indicator] Followed by Redirect to Rotated Phishing Domain · unlock with TL Creds
Indicators of compromise
As of 2026-08-26, Threadlinqs Intelligence tracks 20 indicators of compromise for TL-2026-2150.
Sample indicators (3 of 20)
- Domain:
login[.]microsofte[.]live - Domain:
microcloud[.]homes - Domain:
api[.]keyval[.]org
Defanged, TLP:CLEAR. Full set of 20 indicators: unlock with TL Creds.
Technical analysis
The technique abuses npm/unpkg/npmmirror/Tencent mirrors as free, takedown-resistant hosting and has evolved from a hardcoded typosquat domain to a legitimate key-value service used as a dead-drop resolver.
Security researcher inf0stache first spotted this technique in July 2026 in an npm package named 'china_airlines,' which rendered a fake Cloudflare verification page and redirected visitors to the attacker domain microcloud[.]homes. In August 2026, OX Security identified a much larger, related cluster of 24 npm packages published between August 4 and August 24, 2026 (e.g. bgzxcuite2, prezdentkxheiw, ndmxchdjxn2, @worrisome/reutil) that repeat the same pattern: each package contains almost nothing but a single index.html styled as a Cloudflare Turnstile CAPTCHA or a Microsoft login page, plus a package.json. Installing or importing these packages does not execute code or infect a developer's machine — OX Security explicitly notes the threat actor's use of npm 'isn't to infect developers who install it, but to use the registry and its mirrors as a safe, validated storage for the malware.' The danger is that mirror services (unpkg, npmmirror, Yarn's registry mirror, and Tencent Cloud's npm mirror) expose individual package files directly over HTTP(S), so a URL like https://unpkg.com/ndmxchdjxn2@1.0.0/index.html renders as a live, fully-functional phishing page hosted on a trusted, high-reputation CDN domain — one that bypasses reputation-based web filtering and survives npm registry takedowns because mirror copies of a removed package persist indefinitely.
The obfuscated JavaScript embedded in the HTML does not hardcode its final destination. The earliest observed variant redirected to the typosquatted domain login[.]microsofte[.]live (a fake Microsoft login lure); the current variant instead queries api.keyval.org, a legitimate free key-value storage API, to fetch an attacker-controlled redirect URL at runtime — a dead-drop resolver pattern (MITRE T1102.001) that lets the operators rotate the phishing/redirect destination without republishing or modifying the npm package itself. At the time of OX Security's research the resolved destination pointed to the legitimate ChatGPT website, which the researchers characterize as inert/benign staging rather than the final intended use — they explicitly warn the same infrastructure 'could be weaponized to deliver ClickFix or any other phishing domains.' The explicit reference to ClickFix (MITRE T1204.004, User Execution: Malicious Copy and Paste) signals the intended endgame: using the fake-CAPTCHA page as a lure to get victims to copy-paste and execute an attacker-supplied command, a technique that has driven recent Latrodectus, Lumma Stealer, and NetSupport RAT infections industry-wide.
Severity nuance: GitHub Security Advisory GHSA-35pm-fj6m-m6v6 / OSV MAL-2026-13785, issued for one of the packages (bgzxcuite2, published 2026-08-12), classifies it under CWE-506 (Embedded Malicious Code) and states systems with it installed should be considered 'fully compromised' — OX Security's own analysis contradicts this blanket framing, clarifying that installing the packages is harmless and victims can only be affected after being lured to open the rendered HTML and then following a subsequent ClickFix or credential-phishing flow. This is a live, low-noise campaign (each package saw only 50-300 weekly downloads before removal) that functions as reusable phishing infrastructure rather than a conventional software-supply-chain compromise; it is conceptually related to, but distinct from, the October 2025 'Beamglea' campaign (175 npm packages, documented by Socket) that abused the unpkg CDN in a similar way for credential-harvesting redirects.
MITRE ATT&CK techniques by tactic:
Defense Evasion
T1027 Obfuscated Files or Information; T1036 Masquerading
Initial Access
T1195 Supply Chain Compromise; T1566 Phishing
Execution
Resource Development
Remediation
Immediate actions
- Block or alert on direct browser/user requests to raw .html files served from npm mirror CDNs (unpkg.com, npmmirror.com, the Tencent Cloud npm mirror, Yarn's registry mirror) outside normal package-manager/dependency-resolution traffic
- Add the known redirect/typosquat domains (login[.]microsofte[.]live, microcloud[.]homes) to web and DNS blocklists
- Report the identified packages and their file paths to npm, unpkg, npmmirror, and Tencent Cloud mirror maintainers to request cache purges
Workarounds
- Where dependency workflows do not require it, restrict outbound access to npm registry mirror CDNs (unpkg.com, npmmirror.com) at the network egress layer
Longer-term hardening
- Treat trusted CDN/mirror domains (unpkg.com, npmmirror.com, cdn mirrors of package registries) as potential phishing-hosting infrastructure in URL/web filtering policy, not solely as safe developer resources
- Monitor outbound traffic to free key-value/dead-drop-style APIs (e.g. keyval.org and similar pastebin/config-store services) that can be abused as redirect resolvers
- Run user-awareness training on ClickFix / fake-CAPTCHA social engineering (T1204.004) — legitimate verification pages never ask a user to open a terminal or Run dialog and paste a command
Affected and fixed versions
- npm, Inc. / OpenJS Foundation — npm public registry and its downstream CDN mirrors (unpkg, npmmirror, Tencent Cloud npm mirror, Yarn registry mirror)
Vulnerable versions: 24 malicious packages published 2026-08-04 to 2026-08-24 (e.g. bgzxcuite2@1.0.0, ndmxchdjxn2@1.0.0, @worrisome/reutil); china_airlines (July 2026 precursor variant identified by inf0stache)
Fixed in: Not applicable — mitigated only by npm registry takedown of individual packages; no patch exists for copies already replicated to mirror CDNs, which OX Security confirms persist indefinitely after takedown
Timeline
- Security researcher inf0stache first identifies the npm package 'china_airlines' using a fake Cloudflare verification page that redirects visitors to the attacker-controlled domain locked-ioc-4 — the earliest known variant of this technique.
- Publication window begins for the cluster of 24 malicious npm packages later identified by OX Security (window: 2026-08-04 to 2026-08-24).
- GitHub Security Advisory GHSA-35pm-fj6m-m6v6 and OSV report MAL-2026-13785 are published for the npm package bgzxcuite2@1.0.0, classifying it as embedded malicious code (CWE-506).
- Publication window for the 24-package cluster closes; by this point redirect infrastructure has shifted from the hardcoded typosquat domain locked-ioc-2 to the legitimate key-value API locked-ioc-5 used as a dead-drop resolver.
- The Hacker News and BleepingComputer publish independent coverage of the OX Security findings, contextualizing it alongside the related October 2025 'Beamglea' 175-package npm/unpkg mirror-abuse campaign documented by Socket.
- OX Security publishes research disclosing the full 24-package campaign, the mirror-abuse hosting technique, and its connection to the ClickFix (T1204.004) social-engineering pattern.
- npm removes a subset of the identified packages (e.g. bgzxcuite2, prezdentkxheiw, egair0810, mnteckets, airdzticket, egypt0811, passport811) from the registry; researchers confirm mirror copies on unpkg, npmmirror, and the Tencent Cloud npm mirror remain live and reachable after takedown, while several other packages in the cluster remain live on npm itself.
- Cyber Security News publishes coverage of the campaign, the article that triggered this threat record.
Sources cited
- MAL-2026-13785 — bgzxcuite2 malicious package report osv.dev · Advisory · 2026-08-12
- ClickFix Phishing Pages Discovered in 24 npm Packages ox.security · Research · 2026-08-25
- npm package registry entry — bgzxcuite2@1.0.0 npmjs.com · Vendor report · 2026-08-12
- GHSA-35pm-fj6m-m6v6 — bgzxcuite2 embedded malicious code GitHub · Code repository · 2026-08-12
- 24 Malicious npm Packages cybersecuritynews.com · News · 2026-08-26
- 24 npm Packages Abuse unpkg Mirrors to Host Fake Cloudflare CAPTCHA Pages The Hacker News · News · 2026-08-25
- Hackers abuse npm mirrors to host phishing redirect pages BleepingComputer · News · 2026-08-25
Previous and next in supply chain
Further reading
Drafted by the Threadlinqs research pipeline from 7 sources; published ; last updated —.