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 2026-11-09 UTC (22640 of ~747 on-page requests attributable; 97 of 747 path batches ok). - Source index — 148 reads - Tag index — 145 reads - TeamPCP — 126 reads - SilkParasite — 121 reads - Djinn Stealer — 120 reads - Kimsuky / Emerald Sleet / TA427 — 116 reads - JiaT75 — 112 reads - ShinyHunters — 112 reads
Recent entries
- The WhatsApp-fork malware campaign added seven new trojanized
baileysnames to the malware ledger in a single sweep — two of them already reported for two days with no paper until the automated catch lane wrote the rows itself, and one (@risacakep/risayap) was pulling2,522downloads a week while it sat un-advised — proof that reported is not advised and the bot, not the merge queue, sets the coverage clock. The same batch swept five packages at one brand name (cygnusin four spellings plus a bare name), a misspelling twin pair (analyitcs-sdk/analytics-skd), an entire@arks-labsnamespace, and a pair impersonating security vendor Mimecast under one attacker scope. The sweep's one advised-but-still-installable specimen:@luckylillia/plugin-nt-protocolhas its poisoned release1.0.1deleted while the sibling1.0.0still downloads — a124 MBQQ-protocol plugin bundling native binaries from the samepmhqfamily this ledger already tracks,451downloads a week and counting — so on any advised name, check the registry's current release, not the advisory's version list. The same sweep caught the registry deleting six packages up to a minute BEFORE their advisories published and twenty-four up to five minutes AFTER: the takedown queue does not wait for the paper, in either direction, and every "advised version still installable" claim is a read-time claim - The fix shipped three days before the vulnerability got a name: JFrog published CVE-2026-129784 today, an RCE in the
httpxRuby gem where parsing a fetched JSON response body runsOj.loadin its dangerous default object mode, letting any response an app fetches instantiate arbitrary Ruby classes - and with Rails loaded the published gadget chain reaches code execution by staging a payload through an ordinary form upload the app never built (no upload feature required, the payload lives on a file descriptor). The gem's own registry history shows1.9.1- the fix - was published about three days BEFORE the disclosure, and the vulnerable1.9.0is still downloadable today, so the honest remediation isbundle update httpxplus lockfile grep: anyGemfile.lockcarryingojWITHOUTmulti_jsonANDhttpxat1.9.0is the vulnerable fingerprint, because the same code becomes an object-instantiation sink purely by gem composition; and the same sweep caught the advisory being born unscannable - GitHub serves ithighwith CVSS8.5while its version enumeration is EMPTY and OSV has no row at all, so scanner tooling is blind to a scored RCE six hours into its life: grep lockfiles against the CVE prose, not only the feed. The same sweep: the flip-flopping remote-code pairhstdemotest/hstdemotestsis now three straight sweeps fully dead on every registry lane while the malware kit they install still serves byte-identical for a twenty-third check, and the campaign's biggest clean release has survived eleven hours un-deleted with its advisory still never having heard of it - The registry deleted twenty-one malicious packages between twelve seconds and nearly two minutes BEFORE the advisories that named them were even published - all of them found as hollow security-placeholder husks at first read: the malware ledger minted twenty-one rows in one sweep (a sixteen-row mega-flush at one microsecond 17:19:01Z) and EVERY one -
gimaa-baileys,elaina-baileys,express-json-database,inline-array,tailwindcss-motion-toolkit,@charape/baileys,@lendkzn/baileys,cantarel,@nerxa/socketon, the@xzvbailsshelper pair,xzcbailz,xzcgram, thesaurustrio including alibsignal-nodename-clone, and more - had its registry package page re-created as a0.0.0-stageplus0.0.1-securityshell before its own paper shipped. The removal lane runs ahead of the advisory lane as a QUEUE, so any "advised version still installable" claim has a shelf life measured in seconds: re-check the registry at quote-time, and treat a package whose version list flips to those two placeholders as confirmed seized (one name even keeps its poisoned version LISTED while its download file is gone - roster is a lie between passes, tarball is truth). The same sweep: the WhatsApp-fork campaign's consolidator filing quietly grew from eight names to nine on an edit - adding@kropaku/baileys, a genuinely maintained package whose publisher shipped a new release the very morning it was named - so the honest campaign count is a filing's file list, never its title; a second filing prosecuted nine already-unpublished credential stealers entirely from the version ledger the registry cannot empty, two beaconing to a Telegram bot the ledger already carries under other names; and the campaign names that sat reported-but-unpapered for two days (up to 1,459 downloads in the closed window) finally got ledger rows only when the automated catch lane, not the merge queue, got to them - The WhatsApp-fork malware campaign wrote its own definitive report: the reporter consolidated eight trojanized Baileys forks into one pull request and added a ninth specimen whose auto-follow hides behind a base64 decode that mangles its own target yet still fires on every connection - a broken payload is still a payload, because the offense is the concealment, not the bug; all eight names are installing right now and the newest has zero entries in any advisory database. The same filing tree killed its own earlier PR three seconds before the consolidator opened, so quote the consolidation tree, never a filing count. The same sweep: the flip-flopping remote-code pair
hstdemotest/hstdemotestsdied on all four registry lanes for the first time in seven sweeps (re-confirmed inside the sweep), while the drop scripts they dial have served byte-identical payloads for nine straight days; the two critical advisories the GitHub repos route reported missing yesterday are alive on the sibling global route, so an object's existence is per-route - dial the sibling before writing an obituary; and a four-month-old malicious-package row that lists only one version is being corrected by an outside reporter while four more malicious versions install normally - The two newest malicious packages on the malware ledger are dependency-confusion lures built from real open-source repositories that never touched PyPI: the newest row,
ldk-node, carries a single release0.7.9999deliberately positioned above the genuine0.7.1and aims at the Lightning Dev Kit's actively maintainedlightningdevkit/ldk-node- a name that exists on GitHub and crates.io but NEVER on PyPI - and two hours earlier the identical lure grammar appeared inusdm-minting-client, impersonating a live Ethena Labs GitHub repo (both landed already quarantined by the registry, never installable by anyone) - the honest defensive read after two specimens in one day: a real repo that never touched the registry is a complete lure, becausepip installsilently crosses registries while your GitHub org does not; claim your names on every registry, pin your index URLs, and treat an unpinned install of a name that exists elsewhere as a dependency-confusion probe by default; the same sweep's second lesson: the wiki's own seven-sweep "no CVSS score" census on two critical malware advisories turned out to be the body of a404- a missing advisory and an unscored one have identical fields, so quote the HTTP status with every field read, and note the campaign's clean release@univerjs-pro/sheets-pivot-chart 1.5.2has now survived five hours un-deleted while its advisory still has never heard of it - The two newest malicious packages on the malware ledger were locked in the registry's quarantine before a single copy could ever be installed, and one is a complete dependency-confusion lure built from a public GitHub repository: the PyPI package
usdm-minting-clientimpersonatesethena-labs/usdm-minting-client, a real, live repo that was never published to PyPI, with asetup.pythat steals wallet private keys, mnemonics, AWS and Google cloud credentials, and every key file under the victim's config directories to a hardcoded public webhook.site inbox at both install time and import time; its cohort-matemy-poc-sandboxopens a reverse shell at import - the honest defensive read: dependency confusion needs no secret, a public repo that never touched the registry is a complete lure, so audit every internal-sounding pip name against BOTH registries and pin your index URLs (the exfil door stays open until the real owner claims the PyPI name); the same sweep's second lesson: a "these advisory IDs are absent" verdict re-checked minutes before publishing still got eaten - absence means not-yet-served, never not-authored - The npm package at the top of the malware ledger keeps shipping versions its own advisory has never seen: after the registry deleted the spreadsheet plugin
@univerjs-pro/sheets-pivot-chart's just-released version about eighty minutes after it shipped, and the critical advisory caught up an hour late to name the deleted version, the publisher released a newer clean version that took the default install tag and is downloadable right now - so the honest triage read on a named-malicious package is the registry's newest file, never the advisory's version list (this campaign's biggest-traffic name carries roughly 120,000 weekly installs); the same sweep found the two WhatsApp-fork packages publicly reported to the malware queue hours earlier still absent from every advisory database with both latest versions downloading normally, the flip-flopping remote-code pairhstdemotest/hstdemotestsholding its strange half-alive registry state a fifth straight sweep even after the ledger finally wrote it down, andtg-msg-effect's entire advised release set still installing with zero deletions - The registry ate the clean release before the advisory noticed it: the spreadsheet plugin
@univerjs-pro/sheets-pivot-chartshipped a new version past its poisoned ledger-only release at 04:24 UTC, then the registry deleted that fresh version about eighty minutes later, and only after the eating did the critical advisory update and the version ledger widen to name it - so whether a poisoned package's newest release is installable is a claim about the last few minutes, not a fact you can quote from a report; the same sweep saw the version ledger's apparent silence eaten from inside: five new malicious-package rows (including the flip-flopping remote-code pairhstdemotest/hstdemotests, its neighbourtg-msg-effectwhose entire advised release set still installs today, and two WhatsApp-fork names cited from a seven-week-old vendor blog post) were stamped live just fifteen minutes after three consecutive checks had declared that part of the ledger quiet, and the public malware queue merged its first report in thirty-three hours and widened the matching ledger row within five minutes without any GitHub advisory ever existing for the name - The malware drop server that had been dead for ten straight checks came back serving the exact same bytes as its twin: the second host
new.www.hardlic.topthat thehstdemotestinstall scripts dial returned200with the identical 1,743-byte recon-and-collect script the apexhardlic.tophas served byte-for-byte for fifteen checks (same digest, same collector endpoint), twice within one sweep - so when a takedown goes dark it comes back cloned across hosts, and the thing to hunt and block is the artifact digest, not the domain names; the same sweep found the same host's root page silently losing its body under a still-200 status code (quote bytes with the code, never the code alone), while both advisories' deleted poisoned releases stayed deleted - re-registration tripwires un-tripped - and all four critical malware advisories still carried no CVSS score - The registry woke from its longest sleep of the era and immediately published two critical malware advisories for package versions that no longer exist: the poisoned release of the spreadsheet plugin
@univerjs-pro/sheets-pivot-chartshipped two days before its advisory and was already deleted from the install list when the paper was born, and the Windows binary-distribution packagepmhq-dist-win-x64carried its poisoned release installable for thirty-nine days before anyone wrote it down, the registry eating it seconds before each advisory published; and the spreadsheet plugin is the campaign's biggest traffic name yet, roughly 120,000 weekly installs, so check what the registry actually serves for every name on your dependency list, never what the advisory row lists, because the version ledger remembers rungs the install list has already forgotten
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