2026-11-15T0356Z-intel
One pipeline fire, in full · intel run of 2026-11-15 · sub-agent allocation and telemetry, per-iteration verification verdicts and findings, source-list edits, coverage gaps, bridge invocations, and the run's own verification & coverage notes: what was published, what was dropped at the borderline or judged not relevant (and why), single-source carve-outs, and contradictions. Rendered from runs/2026-11-15/2026-11-15T0356Z-intel.md.
Run telemetry
- Items returned
- 3
- Duration
- 21m 46s
- Tool calls
- 5 WebFetch12 WebSearch95 bridge
- Cited sources
- 2 of 26 in slice
- Items returned
- 3
- Duration
- 20m 30s
- Tool calls
- 4 WebFetch40 WebSearch78 bridge
- Cited sources
- 3 of 21 in slice
- Items returned
- 12
- Duration
- 20m 07s
- Tool calls
- 4 WebFetch7 WebSearch135 bridge
- Cited sources
- 10 of 29 in slice
- Items returned
- 7
- Duration
- 23m 44s
- Tool calls
- 3 WebFetch33 WebSearch110 bridge
- Cited sources
- 2 of 13 in slice
Verification
Deep dive
·
Entries this run published (3) and updated (1)
- Cursor marketplace: a tampered Claude Code extension (versions 2.1.312 and 2.1.313) planted an SSH key, a reverse tunnel and persistence for about eleven hours on 2026-10-29; uninstalling it does not remove the persistence
- PolinRider: a North Korea-linked loader reaches developers through force-pushed forged commits in at least 7,609 GitHub repositories, hijacked npm releases and a fake employer's take-home test, and reads its server address from an Ethereum wallet
- ScreenConnect: Huntress finds 2,223 fake guest sessions named after crypto exchanges on 118 self-hosted servers, enrolled from 2026-11-03 and, in its leading hypothesis, bait for technicians whose clients predate 26.6.5; it saw no malicious use and no overlap with SCKit
- The Gentlemen affiliate: an SSL-VPN account without MFA, a Windows Server 2003 host and a dormant local administrator gave Domain Admin in under an hour; the GPO locker most likely lost a replication race, so antivirus was disabled estate-wide but only one host was encrypted (Ransom-ISAC)
Sources changed (this run)
Edits this run made to sources/sources.json · promotions, demotions, new candidates, and fetch-method / category / reliability / url corrections (the run record's sources_changed[]). Paginated; 10 per page.
5 notes · 2 status · 1 added candidate · 1 consecutive_fetch_failures · 1 last_successful_fetch · 1 consecutive_quiet_periods.
| Source | Change | From → To | Reason |
|---|---|---|---|
| redhat-product-security-cve | status | candidate → active | promotion_due in the state digest: cited by published entries from 3 or more distinct runs |
| github-blog-git | status | candidate → active | promotion_due in the state digest: cited by published entries from 3 or more distinct runs |
| arctic-wolf-labs | added candidate | · → · | First-hand adversary research with its own telemetry; its 2026-11-12 BlueNoroff piece and 2026-11-10 AntiFox alert are independent primaries beside government and vendor reporting; feed parses with dated items |
| connectwise-trust-center | notes | · → · | recipe, GitHub commits API listing works where the HTML advisories answer 403 |
| karimi-honeypot-data | notes | · → · | recipe, GitHub contents and commits APIs replace the tree page |
| msrc-update-guide | notes | · → · | recipe, revised-CVE filter needs isMariner eq false and msrc cvrf reads the full Patch Tuesday document |
| ncsc-ch-focus, ncsc-ch-incidents, bacs-press | notes | · → · | listing recipe confirmed, titles are readable again |
| netzwoche, ransomware-live, fuitesinfos, ransom-isac, databreaches-net, kaspersky-securelist, hadrian-labs | notes | · → · | recipe corrections found by the sub-agents (tag page listing, extract of the country page after the API certificate expired, trailing slash, hrefs through url --direct, feed only, per-post extract resets, second marketing-only sweep) |
| belden-psirt, sygnia, lab52, anssi-cyber-gouv-actualites | consecutive_fetch_failures | · → · | transport failures on every route (no demotion: 403 or anti-bot blocks never demote) |
| (17 slice records) | last_successful_fetch | · → · | fetched and used: last_successful_fetch set to 2026-11-15 and quiet counters reset |
| (68 slice records) | consecutive_quiet_periods | · → · | 200 but nothing publishable in the horizon: counters incremented by one |
Coverage gaps (this run)
Sources this run's brief needed that returned no usable content via any documented recipe. Bridge-recovered or quiet-day sources do NOT appear here. (Distinct from the independent source-accessibility probe at the foot of this section, which probes all active sources regardless of what any run needed.)
| Source (uncovered) | URL tried | Method chain | Status / class | What the agent did instead |
|---|---|---|---|---|
| belden-psirt | https://www.belden.com/security | bridge:extract → webfetch → bridge:jina | 403 transport-403 direct HTTP 403 on extract and WebFetch; the reader pool answered HTTP 402 on every key | none; candidate record, Belden products already appear in earlier PHP2Shell entries |
| sygnia | https://www.sygnia.co/blog/ | bridge:url → bridge:jina | 403 transport-403 direct HTTP 403 (6026 B); the reader proxy relayed the upstream block | none; the record is pinned fetch_method blocked and earlier fires failed the same way |
| lab52 | https://lab52.io/blog/feed/ | bridge:feed → bridge:url-direct → webfetch → jina | n/a transport-timeout connection reset on the feed and url --direct, WebFetch HTTP 503, reader pool exhausted | none; no November 2026 Lab52 post surfaced through any other source |
| anssi-cyber-gouv-actualites | https://www.cert.ssi.gouv.fr/ | bridge:extract → webfetch → bridge:jina | 200 anti-bot-challenge Incapsula challenge on the direct read, jina pool exhausted, WebFetch returned an empty page | WebSearch found ANSSI's 2026-11-09 statement, already carried by the Danfoss entry |
Bridge invocations (this run)
6 bridge calls this run · these are successful bridge fetches (separate from "Coverage gaps" above).
- tools/fetch_source.py cisa-kev and tools/kev_window_diff.py ×1
- tools/fetch_source.py extract ×1
- tools/fetch_source.py url --direct ×1
- tools/fetch_source.py pdf ×1
- tools/fetch_source.py ncsc-csh recent ×1
- tools/fetch_source.py jina (last rung) ×1
Verification findings · all iterations
Per-iteration finding detail. Each table is one verifier pass · what was flagged, how the main agent remediated it, and the outcome. Walking the tables top-to-bottom shows the verifier's debugging trail across iterations.
Iteration #1 NEEDS_FIXES · 17 findings (truth=11, editorial=1, advisory=5) · Claude Sonnet 5.5 · 9m 33s
| F-code | Section | Item · URL/quote | Verifier summary | Remediation · outcome |
|---|---|---|---|---|
| F4 hallucinated-fact | · | Hedge dropped. Ransom-ISAC says "The locker's domain-wide mode most likely failed because of a replication race", "One line in the actor's deployment script most likely lost a race against replication | fixed: title, headline and summary now say 'most likely' for the replication race and for how Domain Admin was reached | |
| F4 hallucinated-fact | · | (low confidence) The only source never names the victim's sector: it says "a Gentlemen affiliate breached a small company" and "The victim's infrastructure is confidential". 'technology' is unsupporte | fixed: sectors emptied, the report names none | |
| F4 hallucinated-fact | · | Both timestamps postdate the files' own last write and the clock when verification started: ScreenConnect file mtime 04:27:47Z, Gentlemen file mtime 04:32:26Z, verifier clock 04:32:53Z at start. The r | fixed: discovered_at of the three new entries set to each file's creation time | |
| F4 hallucinated-fact | · | The two ids do not exist. entries/2026-11-15/ holds screenconnect-crypto-bait-guest-sessions-118-servers.md and gentlemen-affiliate-vpn-no-mfa-da-gpo-locker-failed.md (ls output). Correct the ids in t | fixed: both ids corrected in the run record | |
| F14 ? | · | No cited source states this generalisation for all three routes. JFrog (https://research.jfrog.com/post/accurasofthire-fake-interview/) says the camera-check route hands the candidate a command that d | fixed: narrowed to 'none relying on an npm install script' in body and summary | |
| F14 ? | · | (low confidence) Huntress says "To register, a guest client needs a ScreenConnect server's relay address, which is embedded in that server's guest installer"; it does not say the installer is all that | fixed: body now says a guest client needs only the relay address embedded in the installer | |
| F3 claim-not-supported | · | (low confidence) Sophos (https://www.sophos.com/en-us/blog/malicious-claude-code-extension-distributed-through-cursor-marketplace) says only "Due to coding errors, SSH enablement fails on most macOS a | fixed: summary now attributes the failed log-in to Cursor alone | |
| F3 claim-not-supported | · | (low confidence) Cursor (https://cursor.com/blog/security-incident-update) says "using a stolen credential from the exposed data from our internal system, the attacker placed malicious, unauthorized c | fixed: summary now says 'a stolen credential from the exposed data' | |
| F3 claim-not-supported | · | (low confidence) Apiiro (https://apiiro.com/blog/polinrider-north-korea-worm-forged-github-commits) says "1,949 got injection pushes after their listing date" and, separately, "Where owners cleaned up | fixed: body now says the stolen access 'often still worked', as Apiiro states | |
| F3 claim-not-supported | · | (low confidence, minor) Sonatype says the loaders "locate a transaction sent by the hard-coded wallet"; 'newest' is stated by Apiiro (code comment 'newest tx') and StepSecurity ('most recent transacti | fixed: 'latest outbound transaction' cited to Apiiro, plain HTTP on port 443 to Sonatype | |
| F3 claim-not-supported | · | (low confidence) Huntress: "Across 29 servers with complete connection histories ... the operator revisited the servers every 151 seconds". The summary applies it to all 118 servers; the body is corre | fixed: summary scoped to the 29 servers with complete history | |
| F8 needs-more-research | · | (low confidence) Huntress's first recommendation is "MSPs should upgrade ScreenConnect host clients to 26.6.5 or later" and, if upgrades must wait, remove TransferFiles from roles and session groups. | fixed: takeaway now opens with the 26.6.5 upgrade; the action stays with the CVE-2026-84869 and SCKit entries | |
| F11 editorial-advisory | · | Record-keeping narration about the actions[] field in the record summary, and a composition-rationale sentence in sourcing_note ('the entry follows Cursor'); sourcing_note is also five sentences where | fixed: record summary amended without the actions narration, composition sentence removed from sourcing_note | |
| F11 editorial-advisory | · | The body describes EtherRAT as a Node.js implant that reads its command address from an Ethereum smart contract and the report's own ATT&CK table lists T1059.007 (JavaScript), T1059.001 (PowerShell) a | fixed: T1059.001, T1059.007, T1102.001 and T1568 added | |
| F11 editorial-advisory | · | A single unnamed small-company victim with no nexus stated. The strict bar for incident entries expects the entry to say which ground it clears; the store already carries The Gentlemen entries against | declined: a threat entry with transferable behaviour and detection content that claims no victim nexus; a ground sentence would exceed the word bound | |
| F11 editorial-advisory | · | (low confidence) A claimed client version string used as a hunting key; Huntress itself lists it as 'context indicator, not unique on its own'. Borderline against the zero-IOC rule; consider describin | fixed: guest version string marked as context, not unique, and removed from the action | |
| F11 editorial-advisory | · | (low confidence) The fake-recruiter route is the registry's campaign:contagious-interview playbook (JFrog's post is titled 'Contagious Interview'; the registry already records an overlaps-with edge fr | declined: the playbook is background in JFrog's post; the overlap edge sits on campaign:polinrider and the entity stays unkeyed to avoid a mention-only key |
Iteration #2 NEEDS_FIXES · 6 findings (truth=3, editorial=2, advisory=1) · Claude Sonnet 5.5 · 7m 02s
| F-code | Section | Item · URL/quote | Verifier summary | Remediation · outcome |
|---|---|---|---|---|
| F14 ? | · | (low confidence) Huntress (https://www.huntress.com/blog/fake-crypto-exchange-sessions-screenconnect) says "To register, a guest client needs a ScreenConnect server's relay address, which is embedded | fixed: body now says that to register a guest client needs the server's relay address, which is embedded in that server's guest installer, with no exclusivity claim | |
| F4 hallucinated-fact | · | (low confidence) Huntress hedges: "However, after further investigation, this doesn't seem likely." and "does not match what we would expect from a sandbox"; it does not say it rejects the suggestion. | fixed: body now says Huntress considers the sandbox suggestion unlikely, as its hedge reads | |
| F4 hallucinated-fact | · | (low confidence) Unaddressed half of iteration-1 finding 3: the record was written at 04:44:50Z (file birth time) with completed 04:50:00Z, and the verifier clock read 04:48:42Z and 04:49:53Z when che | fixed: completed and duration_seconds are provisional placeholders that Phase 6 re-stamps from the real end of the fire | |
| F5 missing-citation | · | (low confidence) The citation closes after the first clause. The GPO-first response order is Ransom-ISAC's ("Neutralize the GPO first, before cleaning hosts or rotating anything: 1. Unlink the GPO. 2. | fixed: the unlink-and-delete clause now names Ransom-ISAC and carries its citation | |
| F8 needs-more-research | · | (low confidence) The update adds a second exposure class (Cursor customers whose non-Privacy-Mode agent traces held credentials, and credentials sent to Cursor support) and Cursor's own test for it (" | fixed: Exposure line now carries the second exposure class (non-Privacy-Mode agent traces, notified users) with Cursor cited; main analysis trimmed to stay within the word bound | |
| F11 editorial-advisory | · | (low confidence) Ransom-ISAC's ATT&CK table lists T1490 ("The locker deletes the shadow copies and the change journal as part of its cleanup.") and the post lists shadow-copy and change-journal wiping | fixed: T1490 added to techniques |
Iteration #3 NEEDS_FIXES cap-breach · 4 findings (truth=1, editorial=0, advisory=3) · Claude Sonnet 5.5 · 7m 13s
| F-code | Section | Item · URL/quote | Verifier summary | Remediation · outcome |
|---|---|---|---|---|
| F3 claim-not-supported | · | (low confidence) 'the same wallet' binds back to 'a hard-coded Ethereum wallet' of the Apiiro-cited clause, but the Sonatype page ties GitGuardian's GhostAction report only to its own wallet ('This ex | fixed: reworded to 'the wallet in the hijacked npm releases' so the GitGuardian link attaches to the right wallet | |
| F11 editorial-advisory | · | Ungrammatical clause after the iteration-2 edit (the infinitive 'to register a guest client' has no subject for 'needs'). Huntress: 'To register, a guest client needs a ScreenConnect server's relay ad | fixed: sentence rewritten so 'a guest client' is the subject of 'needs' | |
| F11 editorial-advisory | · | (low confidence) Paraphrase drift: Fireworks says the group is 'aimed at selling tokens from unauthorized access to Large Language Models (LLMs) to illicit routing gateways' and files the cloud-provid | fixed: Update section now words the Fireworks bulletin as it reads (sells tokens from unauthorized access to language models to illicit routing gateways; further activity after a cloud-provider incident) | |
| F11 editorial-advisory | · | (low confidence) A literal claimed client version string, which Huntress lists in its IOC table as 'context indicator, not unique on its own'. Borderline against the zero-IOC rule; the entry already m | declined: Huntress lists the guest version string as a detection field, and the entry already says it is context, not unique on its own |
Verification & coverage notes
The run record's narrative body, verbatim. This is where the run accounts for its own judgement calls: every borderline drop and judged-not-relevant item with its reason, dedup decisions, single-source items and their carve-outs, contradictions, and per-source coverage gaps, so nothing the run considered disappears silently.
Verification & coverage notesrun record body
2026-11-15T0356Z-intel · Sonnet 5.5 · window 28 h · 3 entries published
Verification & coverage notes
Window: gap 25.0 h (daily cadence, previous intel fire 2026-11-14T0255Z), window 28 h; a Saturday-to-Sunday window with the home-region feeds quiet (CERT.at newest 2026-10-21, BACS press 2026-11-05, NCSC Security Hub post 13178 of 2026-11-13). No outage backfill was triggered (the gap is the normal daily cadence); S3 swept the majors' listing pages for posts since 2026-11-13 anyway. KEV sweep: 0 additions in the window, no ransomware-flag flip on a covered CVE. Source slices were ranked on the rotation cursor alone; the previous-two-fires exclusion was not applied because it left the breaches slice empty.
Published: three new entries (PolinRider, ScreenConnect crypto-exchange bait sessions, The Gentlemen affiliate report) and one update record on the Cursor marketplace entry (Cursor's own report of 2026-11-13 gives the origin the entry said nobody had published). Deep dive: none, nothing cleared the bar.
- Single-source: 2026-11-15/screenconnect-crypto-bait-guest-sessions-118-servers (Huntress alone, hypothesis stated as unobserved, ConnectWise bulletin unreadable); 2026-11-15/gentlemen-affiliate-vpn-no-mfa-da-gpo-locker-failed (Ransom-ISAC, reliability C, no victim named).
- borderline-drop: jDownloads CVE-2026-117551 (Joomla), needs upload permission, no exploitation, Swiss footprint unknown.
- borderline-drop: Fortinet FG-IR-26-187 revision of 2026-11-13 (Brotli decoder flaw), product-status list only, no fix or workaround, advisory not cited by the covered entry, so no patch decision changes.
- borderline-drop: Qubole Wiz update of 2026-11-11; the "n0n" claim is relayed from a newsletter and unverified, the separate key-testing is already in the entry's sourcing note, no deployment of the bootable image is known.
- borderline-drop: GeoVision camera software abuse (Huntress, 13 organisations), initial access unnamed, no CVE, no region; dropped on 2026-11-14 for the same reason.
- borderline-drop: Unit 42 CL-STA-1238 (Cairn) and Elastic, Talos and Arctic Wolf agentic-framework internals (ARTEX, AntiFox), US-centred targets, class already covered by the ARTEX and CyberStrikeAI entries; dropped on 2026-11-13 and 2026-11-14.
- borderline-drop: FBI FLASH and DOJ action on FaceStealer, Swiss nexus indirect; dropped on 2026-11-13 and 2026-11-14, no new fact.
- borderline-drop: Project Zero "Full X" policy, effective 2027, no near-term decision.
- borderline-drop: Elastic SANSAR (Ray job agent), CloudSEK Silent Lynx (Central Asia), Recorded Future survey (vendor survey), AMD SEV-SNP Mior and Skyfall (hypervisor-level attacker, CVSS 5.9).
- borderline-drop: React Server Components CVE-2026-126187, application-code preconditions, no exploitation or proof of concept; dropped on 2026-11-12 and 2026-11-13.
- out-of-window: Saxony emergency pager exposure (heise 2026-11-12, Spiegel 2026-11-11) and the Operation Endgame round (Europol 2026-11-12), window_hours=28; both were also dropped by the 2026-11-13 and 2026-11-14 fires, the pager item for lack of an access vector or actor.
- borderline-drop: Quantum Systems QBase update-server compromise (sources 2026-11-13, no Swiss customer), Mittwald MCP-server intrusion (German host, post-mortem pending), Fireworks AI incident (folded into the Cursor update), Microsoft 365 outage of 2026-11-12 and 2026-11-13 (capacity upgrade, not an attack).
- Backlog: the one open row (ChimeraZ claim against the Belgian rescue zone Hainaut-Est) struck on 2026-11-15: six French-language searches found no confirmation and the cited page does not mention the zone.
- Coverage gaps: belden-psirt (403 on every transport); sygnia (403 on every transport); lab52 (connection resets, reader pool dead); anssi-cyber-gouv-actualites (Incapsula challenge); databreaches-net article pages (403, feed works); fbi-gov, quantum-systems.com, beobachter.ch and connectwise-trust-center HTML pages (403, other routes used); jina reader pool exhausted.
- Watchlist: none configured, sweeps were no-ops.
- Candidate sources considered and not added: norforge (no feed), fireworks-ai-security-bulletins (low volume, the Cursor update carries the bulletin), lemagit-cyberhebdo (weekly issue URLs change each week).
- Mid-run stop-hook nudges to commit were declined; the run commits once, after the gate and the verifier loop.
- Verification: three iterations, all NEEDS_FIXES (iteration 1: 11 truth, 1 editorial; iteration 2: 3 truth, 2 editorial; iteration 3: 1 truth, 0 editorial, a low-confidence wallet-attribution wording). The final residual of 1 was fixed after the pass and not re-verified; the loop stopped on the low-residual early exit. Iteration 1 caught two timestamps written ahead of the clock and a dropped hedge, both fixed.
- The gate's dedup advisory on actor:thegentlemen is deliberate: the 2026-11-07 Kaspersky entry covers a different incident (a monitoring server's root password on ESXi and Proxmox), this one an SSL-VPN intrusion with a GPO locker failure.
- Note for the next audit: the Cursor update was trimmed twice to stay within the 550-word analysis bound; the entry carries its original wording otherwise.
← Operations dashboard · day page 2026-11-15 · run-record contract: docs/pipeline.md