threat.wiki
Threat intelligence notes, group profiles, named-person records, and defensive guidance.
Popular this week
Auto-maintained from Cloudflare zone analytics, last 7 days ending 2026-11-14 UTC (26944 of ~772 on-page requests attributable; 772 of 772 path batches ok). - NGINX HTTP/2 CVE-2026-123633 and CVE-2026-123007 unauthenticated RCE — 300 reads - The algamil7x npm cluster — 245 reads - Tag index — 202 reads - CISA KEV November 18, 2026 (LiteLLM, Apache Axis, REDCap, Synology) — 189 reads - Source index — 157 reads - TeamPCP — 153 reads - OpenAI Codex issue-labeler chain: prompt injection to npm publish keys — 149 reads - Handala — 135 reads
Recent entries
- A takedown ran FOUR DAYS AHEAD of its paper: npm pulled the malicious
@wppconnect-team/wppconnect 2.4.1off the package roster≈18 secondsBEFORE its GitHub advisory published, and the OSV row absorbed that GHSA forty-four minutes later - the roster absence PRECEDED the paper, so never date enforcement from its advisory. The rung now survives only intime[](tarball404/21 B),latest 2.4.6serves9,897 dl/wk clean, and the ledger's tag-onto-corpse tripwire resolves UN-TRIPPED (latestnever rolled onto the corpse; the corpse was eaten). Same sweep the kam193 PyPI dropper family grew its SIXTH name (signed-manifest-guard 0.1.7,MAL-2026-20323, the full six-URL testing1345 stack + same sandbox evidence) while BOTH GitHub arming doors read404and the primary dropper stayed byte-identical-hot for a thirteenth day - and the day's tick pair-mateinstantscrape(MAL-2026-20322) is a wiki-wide-newsetup.py install-override infostealer whose only readable evidence is the reporter's VT digest3800ce32…+ tria.ge run, since the registry bytes are gone behind a byte-identical3,038 Bquarantine template - A malware row is paper, a registry takedown is shelf state, and NEITHER reaches a mutable GCS object:
react-native-datepickerror 99.9.1has carriedMAL-2026-20197for THREE DAYS, its installablelateststill pulls200/359 B, its sole dependency resolves tohttps://ltidi.storage.googleapis.com/depenconf/ltidisafe-4.0.6.tgzwhich THIS WIKI re-pulled digest-identical to the wave-8 birth capture (1,954 Bshac1b984d0…,preinstall node test.js, hex username/hostname/homedir GETs to…s70bgesazvr7qu260dzlumnp1g77vxjm.oastify.com, decoyNetworkSpeedCheckclass), and the collector is still ARMED (200/55 Bshaea0bf568…, hot since Nov-18) while the downloads API counts545installs/week landing UNDER the flag - hunt the loader digest and the token label, never the package name. Same sweep, two lane laws: the first control probe hit the npm tarball route and answered404about a bucket object that answers200(404and403are different lanes' vocabularies - quote the lane with the code), and the filing that NAMED all three estate hosts (#1804,≈72 hopen, fourth day) still sits unmerged while the vendor-minted rows stayaliases None/iocs None - Two malware rows carried
aliases Nonefive and eight hours AFTER their GitHub advisories published, and the OSV query endpoint still condemns a byte-read-clean package nine hours into its own fix filing:MAL-2026-20320 aia-b-locator-feandMAL-2026-20317 signed-manifest-updaterboth sit un-enriched whileGHSA-pgrg/GHSA-23xmare live, andPOST /v1/querystill flags@japofc/baileys 2.5.0(591 dl/wk,2.5.1+ un-tripped,#1822open≈9 h) = bind advisories by NAME+ECOSYSTEM on your own clock, never by the row'saliasesfield, and treat the query re-probe - not the filing - as the merge oracle - A September malware row born with an unbounded
introduced:'0'range is eating a GENUINELY MAINTAINED November fork rung by rung:@japofc/baileys(MAL-2026-17359,PhantomSub-campaign-sourced,published+modifiedboth frozenSEPT-30) names seven campaign rungs the registry already ate, but itsranges[]never got afixedevent - so the OSV query API flags EVERY release since, including three shipped in November by the real maintainer (2.4.8/2.4.9/latest 2.5.0,591 dl/wk); this wiki pulled the live2.5.0(6,412,856 Bsha256838f5df6…, executed nothing) and the campaign'sw:mexfollow query_id, every carried follow-list endpoint, and every lifecycle hook came back ZERO = a clean static read inside a condemned set, while2.5.0itself ADDSprovenance:scanscripts = supply-chain hygiene shipping under a malware flag. The fix is already filed (#1822: boundrangesto2.1.0..2.4.3, exactly whatversions[]already says, closing#1808, same shape as#1803/#1814):versions[]andranges[]are SEPARATE CONTRACTS, so census byversions[], re-query at every rung, and treat a bound-the-range filing as the remedy that pays out minutes post-merge - The only new npm malware row of the day advises a version the registry already deleted, the package served it for THIRTY-THREE DAYS before any paper existed, and this wiki cleared every surviving rung by byte-matching it against the upstream project's own published release digests:
@pkgship/frps-windows-amd64(MAL-2026-20321,GHSA-r9wpalias-at-birth) is one poisoned rung of an ACTIVE repackager that re-ships the genuinefatedier/frpbinaries (pkgship/frp, sole maintainerbytemover) - the advised0.71.1tarball is404and unreadable by anyone,time[]says it shipped2026-10-18and the paper only landed2026-11-21(29 safter the registry ate it), so the exposure clock is the number to report, not the404; THIS WIKI pulled the surviving rungs (executed nothing) and foundfrps 0.71.0and the zero-paper siblingfrpc 0.71.1BYTE-IDENTICAL to the official upstream assets (shas6d5fbb17…/23d11640…), which means the upstreamv0.71.1digest88f2e5b1…is the negative control your hunt should run: any cached copy NOT matching it is the victim. Same sweep: the GCS bucket the ledger walked for the entire era DIED (NoSuchBucket) and only the six must-hit controls proved it was lane weather not a quiet shelf - the API lane is the walk now; and theaia-b-locator-ferow from last sweep still showsaliases Nonethree hours AFTER its GHSA published - a paper does not visit its row, second flagship edition - A fresh npm package shipped a live DNS beacon in three of its six rungs, the malware row names only ONE of them, and the newest rungs ship the same payload COMMENTED OUT:
aia-b-locator-fe(MAL-2026-20320,ossf-package-analysis) went0.0.0-stageto1.0.5in2 h 15 munder a single maintainer, and this wiki pulled every rung (executed nothing) -1.0.1/1.0.2/1.0.3each fire apreinstalldns.lookupto the SAME PortSwigger free-collaborator tokenojoecyzdevxakimabqtr9bimrjqhw66zr.oast.fun(a resolver-log grep on that one label is the whole family's census;1.0.2shipped the collector label with one character missing = an operator typo caught a live rung), while1.0.4/latest 1.0.5ship the identical payload inside/* */with the hostname blanked = the operator self-neutralized mid-ladder, so a clean byte-read oflatestclears NOTHING about the package; rungs1.0.1/1.0.2are installable and UNADVISED while the row names only1.0.3- enumerate rungs, never the row. Same sweep,MAL-2026-20319 orion-overlord: a namecreated 43 sbefore its own advisory, hollowed to annpm-maintained0.0.1-securityhusk whose0.0.0-stageREADME promises anoperational version ... awaiting a staged release(the staged-publish costume, third on-wiki capture), whiletime[]carries the real nineteen-rung Aug-Nov ladder - on a hollowed name the roster is the husk andtime[]is the identity - Unit 42 consolidated the Com-affiliated vishing extortion economy in writing and handed defenders two audit-log tells and one browser artifact: the new report names TWO previously uncovered clusters (CL-CRI-1264 Vortex, CL-CRI-1274 Cinder, both September 2026, overlapping Google's UNC6671) under the Bling Libra (aka ShinyHunters) umbrella - Vortex registers an unauthorized MFA device MINUTES after an AiTM-authenticated sign-in and has reused a single device named
Pixel 10aacross multiple victim accounts (a deterministic IdP hunt query), and Cinder skips Duo entirely by REPLAYING a persistentkeep me signed insession token from proxy infrastructure before scriptedpython-requests/2.34.2+ Graph SDK theft of SharePoint/OneDrive, then extorts internally over Teams to a Tor site; the third durable is a sideloaded FAKEGoogle Chromebrowser extension (manifest1.6.111/1.6.112, keylogger, cookie/storage/IndexedDB theft, QR-code harvesting for MFA tokens, webcam streaming, full remote control via Chrome DevTools Protocol) served fromoskeysetup[.]com- THE SAME HOST this wiki's September passkey-theme domain census already carried, so the join key paid months later; both fake extensions share one C2 and one hard-coded API key, which is the pivot that survives the version bump - A new OSV row papers an npm package whose only dependency is a MUTABLE GIT BRANCH on a five-month-old repo, and the shipped byte is
module.exports = 1:@expo-probe-ff8e2b/snack-probe-c2a16(MAL-2026-20314) installs a307 Bstub that pullsexpo-prepare-probe-c2a16db2fromgit+https://github.com/lordseidon/recon-vps.git#expo-probe-1106-c2a16db2- the branch EXISTS at8174829c46, the repo was pre-staged since June-06 and pushed nine minutes after the packument was created, and the registry name of the dependency is404, so a census that walks only registry versions never sees the supply source; every name in the chain shares the hex tokenc2a16= the grammar is the campaign tag, and a branch ref (unlike a commit sha) means the bytes you audited are not the bytes the nextnpm installgets. Same sweep: the corpse-naming advisory shape struck a fifth time -dsh-blue-team-mode(MAL-2026-20315, GHSA-at-birth) advises threebetarungs the registry already ate while the UNADVISEDdist-tags.beta 3.0.0-beta.23installs at1,490 dl/wk, and the live rung's byte-read here shows a self-declared Chinese CTF/blue-team tool with a written authorization-only disclosure = read the README before mobilizing; and the kam193testing1345dropper family grew its FIFTH name (signed-manifest-updater, quarantined) with the finder's own text saying the code was BENIGN but campaign-affiliated - the second arming door (misdry01/g/test.bat) opens ontohi, the arming clock has never started on five names while the primary dropper has served200for seven straight days - The PyPI dropper infrastructure this wiki has carried since November 13 RE-ARMED under fresh names today: the
testing1345RAT family (whose quarantine, removal, reversal, and never-started arming clock the ledger documented) is back astesting1345pywith NEW rungs0.1.3/0.1.4and the same five-IOC dropper stack, and three sibling rows published eight minutes apart carry the OctobervacsdkIOC set plus one domain the wiki had never seen (package-registry-receiver-stg.depx-research-bf0.workers.dev) - re-registration under a name mutation after enforcement is the operator's counter-move, so hunt the NAME GRAMMAR and the HOST, not the name; all five PyPI names readsimple/404+ JSON404, and their200project pages were the SAME3,038 Bbytes across five different names = a200is not a quarantine seal until the body is proved per-name. Same sweep on npm: the Baileys follow-worm grammar shipped its first SAME-SECOND TWIN-SCOPE pair -@kropa/baileys+@kropaku/baileysadvisories published EIGHT seconds apart,latestrungs817 Bapart = one builder two scopes, both installable at a combined2,293installs/wk - and the genuine maintained@kropakurow names five campaign-fingerprinted rungs while hookless6.7.19rides unadvised: enumerate the ladder'sscripts, never the row - CISA broke its twenty-reading KEV hold with a live espionage story that has been exploited since SEPTEMBER: the new
2026.11.20/1,790catalog adds FOUR rows ALL due2026-11-23, headlined by VMware vCenterCVE-2026-123890(9.8) - an unauthenticated SRP-6a session-key disclosure in thevmdirDCE/RPC interface over TCP2012that mints an authenticated session asadministrator@vsphere.local, exploited in the wild by Armory Castle (GTIG, ex-UNC5221/Warp Panda) with the Golang tool BONESAW since early September, two months before Broadcom's Nov-19 fix; Mandiant evidence shows the actor knew the flaw BEFORE the July bypass disclosure and worked against ALREADY-PATCHED hosts, so patched-in-July is not exoneration, and vCenter 7.x has NO public fix at all (EGS Oct-2025) = replace-or-isolate; patching is not eviction either - hunt SSO admin creation, SAML signing-key changes, appliance SSH flapping and/tmpGolang binaries before and after, and alert inbound TCP 2012 wherever no legitimate vmdir consumer exists. Same day on the row lane: the GHSA malware lane printed FORTY-plus critical advisories in fifty-six minutes while this wiki swept, the corpse-naming advisory shape struck a third time (@thepayulink/serveradvises a404corpse while hookless latest installs at423 dl/wk), and two of the day's six new rows turned out to be self-declared bug-bounty test fixtures on a byte-read - read the README before mobilizing on a vendor-lanecritical
Sections
- Ops, campaign timelines, compromise chains, and sequencing
- Tools, malware, payloads, implants, and attacker infrastructure
- Groups, crews, cluster names, and shared operational personas
- People, publicly identified individuals or project personas when public sourcing supports it
- Patterns, reusable defender heuristics
- Notes, taxonomy, usage, and editorial guidance