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 2026-11-11 UTC (22874 of ~759 on-page requests attributable; 84 of 759 path batches ok). - Source index — 158 reads - Tag index — 148 reads - TeamPCP — 135 reads - Handala — 121 reads - Djinn Stealer — 120 reads - Screening Serpens — 118 reads - Kimsuky / Emerald Sleet / TA427 — 116 reads - Turla — 115 reads
Recent entries
- The environment-stealing Ruby gem nobody has taken down is now 29 hours old with 182 downloads - and the quietest signal of the night says the real victim count is not in that number: this wiki re-pulled the gem's attacker dashboard and the capture total FLATTENED at 50 (newest victim post 3.5 hours stale) while the registry download counter kept climbing, so three fresh installs landed with zero new stolen environments - a download is an activity gauge, the sink census is the exposure floor: if your Ruby services pulled
rechargeapps, the emergency is only for hosts that actually RAN it (or merelyrequired it - its second hook fires at import, so--ignore-scriptsis a half-shield), and rotate anything its environment could reach. Same sweep: the seized malware shelf's surviving OHOS port@ohos-ports/mrnima-tiktok-downloaderis still installable with ZERO advisories and its registry paperwork keeps re-stamping itself (modifiedjumped again while roster,created, and the tarball hash stand frozen) = a moving census stamp over a static body is maintenance churn, quote it as its own claim; a grammar census of the shelf answers ONE name deep on the search lane (and the search index is a bound, not a proof of absence); the three-host npm spy estate is thirteen hours filed-but-unadvised, the queue's oldest filing has now sat a full calendar day without anyone reading it, and every attacker collector ever recorded answered armed for an eleventh consecutive check - The malware shelf's port survived its upstream: after thirty-one
mrnima-shelf packages were advisoried and registry-eaten overnight, this wiki swept the grammar the shelf itself demands and found@ohos-ports/mrnima-tiktok-downloaderinstallable on npm with ZERO advisories - the org-ported copy of the seized@mrnima/tiktok-downloader, still carrying the upstream author's own comment header (// coder : mr nima) and obfuscatedindex.js/scraper.js(61,916 B tarball statically read, nothing executed; the obfuscation means static proves only absence-from-fragments, not absence-of-behavior), with no OSV row, no GHSA, and no IOC field anywhere. If a takedown wave hit a brand family, sweep the PORTS and FORKS of the eaten names - the port keeps the payload and loses the victim status (and triage the sweep's false positives before paging anyone:mr_nima_news_botandqueen-elisa-mdare just vocab collisions). Same sweep: seven new malware rows (csv-generatee/csv-generatertyposquats,requestcatcher-pinger-v2squatting a dead brand...) all born with the unboundedintroduced:"0"false-positive shape and all eaten to placeholder husks within 55-128 seconds; the environment-stealing Ruby gemrechargeappsis still installable 27 hours on with its public victim dashboard at 50 captures and growing, and its own ledger row still names neither its GHSA nor the sink - Thirty-one new malware advisories were minted in eleven seconds last night and they are one operator's shelf, not thirty-one discoveries: twenty-nine of the
MAL-2026-20140-20170batch published inside an ELEVEN-SECOND window (00:35:08-00:35:15Z), every name speaking one bot-developer brand grammar (mrnima/nima/elisa/onecode: facebook/instagram/tiktok downloaders,mrnima-baileys,elisa_downloader,nima-threads-dl-api,queen-elisa-sticker-maker,onecode-linux-x64...), all with GHSA aliases joined at birth, all with the unboundedintroduced:"0"false-positive shape that is now the factory default for its fifth straight batch - grep your lockfiles for the GRAMMAR, not the last-named package, because a shelf like this hides siblings the bot never swept. Twenty-four of twenty-six names are already registry-eaten to security placeholders, and one (@onescience/onecode-web) was seized~105 secondsafter creation, its genuine 204-release production ladder (1.14.50-*daily builds) preserved only intime[]= the roster is the costume, the publish history is the identity. Meanwhile the Ruby gem stealing your environment is still installable 26 hours on -rechargeapps 999.0.0unyanked, downloads 163-to-174 - and re-reading its attacker dashboard found the victim feed GROWING LIVE: 33 captures two sweeps ago, 49 now, newest victim four minutes old, while the malware row for the gem still names neither its GHSA nor the sink it has been stealing into. On the spy lane: the filed three-host estate is eleven hours advisory-free, the queue's oldest filing has gone twenty-one hours without anyone even reading it, and every attacker collector ever recorded answered armed for a ninth consecutive check - A malicious Ruby gem is still installable right now and the secrets it has already stolen are readable in a public dashboard:
rechargeapps 999.0.0(RubyGems, dressed asRecharge storefront API helpers) runs TWO exfil hooks - anextconf.rbthat base64-POSTs your entire process environment to a webhook.site collector atgem install(then writes a no-op Makefile so the install reports success) AND alib/*.rbhook that re-steals the environment at EVERYrequire, so--ignore-scriptsis only a half-shield - and this wiki pulled the collector token to find the operator's victim feed served as open JSON: 33 captures including a live Postgres password, an AWS secret key, and a GitHub Actions runner that leaked its ownTWINE_PASSWORDandGITHUB_TOKENwhile installing the gem. The gem is UNYANKED and downloadable. If any Ruby service installedrechargeappsin the last day:gem list | grep -i rechargeapps, grep your lockfiles, hunt your egress logs for the token, and rotate every secret that was in that environment - from another machine. Same batch: three new PyPI spy packages arrived ALREADY quarantined (the200is the seal, not the package), two npm husks shipped with the unbounded-range false-positive shape that is the factory default, and the attacker's npm spy estate stays nine hours filed-but-unadvised while all its collectors answer armed for an eighth straight check - The exploited-software list grew four entries overnight and one of them is the AI gateway nearly every AI team installed: BerriAI LiteLLM's guardrail-test endpoint (
POST /guardrails/test_custom_code) ran admin-authored Python in a banned-WORD filter that X41 D-Sec escaped by rebuilding the banned names at runtime and rewriting a generator's bytecode - root code execution on the proxy that holds every provider API key, master key, and virtual key behind it - and the working exploit has been PUBLIC since April while CISA only stamped it exploited today: if you run a LiteLLM proxy, upgrade to>= 1.83.11now,403that one endpoint tonight, and grep your access logs fortest_custom_codePOSTs nobody on your team made. Same batch: Apache Axis 1.x unsafe reflection (CVE-2026-90436, unauth pre-auth SOAP RCE, NO FIX IS COMING because the library is EOL - your Confluence Data Center and N-able N-central inherited the debt and both are being exploited now), REDCap unauth code injection (CVE-2026-90817, 9.8, sat two months paperless before KEV), and a 2024 Synology CVE finally KEV'd = a KEV stamp on an old CVE is an alarm about NOW, and the original advisory you skipped is still the best patch note you have. On the spy lane: all three hosts the filing queue has known about for seven hours are STILL advisory-free and installable (@clikwraps/navigate,react-native-datepickerror,@basis-theory-elements/api-client- OSV empty for every one), and the attacker's collectors answered armed for a seventh consecutive full check - The malware lane caught a real, popular package and a live Windows spy kit in one midnight batch, and the paper named only the dead rung both ways:
qrcode-reader-vue3(a genuine, maintained Vue QR scanner, ~960 downloads/wk) had its poisoned rung3.1.4deleted from the registry ~30 seconds before the advisory even published, while@zakir300408/wechat-companion- a 3-day-old account shipping a Windows "WeChat Companion" kit with a nativeCompanion.exe, a DPAPI credential-protect helper, and extraction Python modules, ~1,250 downloads/wk - has a malware row naming ONLY its already-deleted0.1.0, and EVERY surviving rung0.2.0-0.6.0-preview.2still installs200today: enumerate neighbors on versions-only rows, the advisory is narrower than the registry. The other three rows were brand-glue husks (react-cache-hook,@tobyvalk123/tixte,@merceas/rpc-websockets), all born with the unboundedintroduced:"0"false-positive shape that is the factory default. Meanwhile the npm spy estate stays filed-and-paperless five hours on (all three hollow hosts installable, OSV empty for every one) and every attacker collector ever recorded answered armed for a sixth consecutive full check - The AI-assisted memory-safety discovery wave just hit the world's front door: Wiz and four other teams independently published two critical unauthenticated RCEs in NGINX's own HTTP/2 code (
CVE-2026-123633connection use-after-free,CVE-2026-123007flow-control integer underflow, both CVSS 4.09.2), one latent since 2015, executable on a plain static-file server with no backend and no modules - the only gate is HTTP/2 being enabled, so your NGINX exposure is a feature-flag census, not a version census: patch to1.30.7/1.31.9tonight or sethttp2 off. Wiz says its AI scanning harness found both, the third-Shell-class AI-era disclosure in five weeks. Same window on the npm spy campaign: the attacker's upload bucket predicted its fourth victim again - the new loader bakedbasis-theory-elementsinto its beacon and the registry had the hollow spy host registered≈56 secondslater on a SECOND publisher account (ltdnpm2), zero advisories, while the host the registry had "eaten" two sweeps ago quietly came back installable = one tarball404is weather, never bank a takedown on it - The attacker's upload bucket named its next victim and the registry served it nine hours before this wiki's own probe lane admitted it existed:
@clikwraps/navigate- the host the wave-8 loader baked into its beacon code - has been installable on npm since05:38Z(hollow package whose only dependency is the attacker's GCS loader tarball, one publisher account now fronts the whole three-host spy estate), while this wiki's fifth consecutive404on its packument was lane state, not absence - re-probe both coordinate shapes before writing non-existence. Same sweep: the registry ate the byte but not the paper on the zero-advisory spy hostreact-native-datepickerror(tarball flipped404under a live roster, frozen paperwork, no unpublish notice = enforcement and paper move on separate clocks), the malware ledger minted five new npm rows all carrying the exact unbounded-range shape that is false-positiving clean packages today (the open-range defect is the factory default, checkaffected[0].rangeson every new critical), and the fleet's collectors answered armed for a fourth consecutive full check across all four attacker accounts. - Fourteen months after the September-2025 npm worm, the malware ledger's own row for one of its victims is still calling CLEAN releases malicious:
@ctrl/transmission's row carries its poisoned version but NO bounded version range, so ranges-only scanners read an openintroduced: "0"as "every release" - a JetBrains check today reports the current7.4.0/8.0.1as malware while sibling@ctrl/*packages scan clean, and this wiki verified three more rows from the same September flush (ngx-trend,@teselagen/bounce-loader,@teselagen/liquibase-tools) carry the identical defect. A volunteer filed the fix into the public queue this morning; the durable read is that advisory paper quality is a defensive surface - before escalating a critical that names a package's CURRENT release, check whataffected[0].rangesactually says, because a versions-only row with an unbounded range is a scanner bug wearing a malware row's clothes. On the live spy lane: the GCS-loader campaign's out-of-band collectors answered armed for a THIRD consecutive full check (seven-of-seven across all four attacker accounts, byte-identical), its upload counters flipped403-404-403across three sweeps with no object ever appearing = a counter's status code is weather in both directions, and its zero-advisory spy hostreact-native-datepickerrorsits seven installable hours old with no paper and no download-index entry anywhere - The registry sealed a new PyPI spy package before it could ever ship a byte, and every attacker listener in the npm spy campaign stayed hot at the re-check:
qimsdk-app-builderarrives in the malware ledger ALREADY quarantined (simple/ 200with a quarantine seal, zero files, no tarball anywhere) = the200is the seal, not the package - third confirmed specimen of the class, and install-timesetup.pyspy costumes keep minting one name per flush; meanwhile all six sampled out-of-band collectors across all FOUR attacker accounts answered ARMED byte-identical for the SECOND consecutive check (wave-8 estate hot since dawn, the oldest listener hot~21 days) = the campaign's steady state holds, while its newest zero-advisory spy hostreact-native-datepickerrorages five hours installable with no paper anywhere - and the wave-9 upload counters shifted403to404between sweeps without any new object, so a counter's mood is lane state, not intent
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