CTIPilot
AI-generated · no human review · verify critical claims against the linked source. how it works →

2026-11-19T0355Z-intel

One pipeline fire, in full · intel run of 2026-11-19 · 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/2026-11-19/2026-11-19T0355Z-intel.md.

Run telemetry

2026-11-19T0355Z-intel intel prompt v4.25 publish ok
5h 02m duration 6 published 8 updates
Claude Sonnet 5.5 (claude-sonnet-5-5) main agent
S1 Claude Sonnet 5.5 (claude-sonnet-5-5)
Items returned
21
Duration
38m 01s
Tool calls
7 WebFetch18 WebSearch215 bridge
Cited sources
3 of 21 in slice
S2 Claude Sonnet 5.5 (claude-sonnet-5-5)
Items returned
6
Duration
22m 53s
Tool calls
4 WebFetch22 WebSearch85 bridge
Cited sources
4 of 30 in slice
S3 Claude Sonnet 5.5 (claude-sonnet-5-5)
Items returned
9
Duration
25m 59s
Tool calls
5 WebFetch8 WebSearch95 bridge
Cited sources
2 of 15 in slice
S4 Claude Sonnet 5.5 (claude-sonnet-5-5)
Items returned
5
Duration
24m 30s
Tool calls
6 WebFetch32 WebSearch95 bridge
Cited sources
0 of 10 in slice

Verification

unconfirmed CLEAN · waived: single CLEAN at iteration cap (iteration 8 post-fix pass after the iteration-7 N #1 NEEDS_FIXES · Sonnet 5.5 · t=17 e=17 a=7 #2 NEEDS_FIXES · Sonnet 5.5 · t=14 e=4 a=4 #3 NEEDS_FIXES · Sonnet 5.5 · t=7 e=8 a=2 #4 NEEDS_FIXES · Sonnet 5.5 · t=8 e=5 a=3 #5 NEEDS_FIXES · Sonnet 5.5 · t=3 e=1 a=1 #6 CLEAN · Sonnet 5.5 · t=0 e=0 a=5 #7 NEEDS_FIXES · Sonnet 5.5 · t=3 e=1 a=4 #8 CLEAN · Sonnet 5.5 · t=0 e=0 a=3

Deep dive

·

Entries this run published (6) and updated (8)

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.

3 added candidate · 1 promoted to active · 1 fetch_method · 1 health_cmd · 1 notes · 1 last_successful_fetch · 1 consecutive_quiet_periods.

SourceChangeFrom → ToReason
nginx-security-advisoriesadded candidate· → ·project primary with per-CVE ranges, the only readable F5 or nginx source for the nginx2root entry
unit42-timely-threat-inteladded candidate· → ·dated plain-text exploitation notes; its REDCap note predated the CISA listing by six weeks
polizeiticker-chadded candidate· → ·RSS relay of cantonal police communiques that carried the Basel-Landschaft incident while baselland.ch refused every transport
mitel-security-advisoriespromoted to active· → ·three contributing runs (sources.promotion_due)
sygniafetch_methodblocked → webfetchWebFetch with the outbound-links template now reads the listing; failure counter reset
typo3-security-advisories, cnil-frhealth_cmd· → ·source_health UNSOLVED list: the landing page carries no security vocabulary (typo3, recipe is the feed) and the 503 origin fell through to the metered reader (cnil, now url --direct); not re-probed, source_health has no single-source mode
lab52, belden-psirtnotes· → ·no transport reads lab52; Belden advisories read as PDFs on assets.belden.com
cisa-kev, enisa-euvd, fortinet-psirt, advisories-ncsc-nl, cert-at, cert-eu, ncsc-ch-security-hub, zscaler-threatlabz, cyderes-howler-celllast_successful_fetch· → ·fetched and used by an entry or a changelog record published this run
(64 slice records)consecutive_quiet_periods· → ·200 but nothing publishable in the horizon: counters incremented by one; the 9 sources used were reset to 0

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 triedMethod chainStatus / classWhat the agent did instead
lab52https://lab52.io/blog/bridge:feed → bridge:extract → webfetch → bridge:jinan/a connection-reset
every direct transport reset the connection, WebFetch returned 503, the reader pool is dead, Wayback TLS/429
a WebSearch cache shows the newest post is 2026-10-05, so nothing inside the lookback is known to be missed
belden-psirthttps://www.belden.com/securitybridge:extract → webfetch → bridge:jina403 transport-403
HTML listing 403 on every transport, reader pool dead (402)
advisory PDFs on assets.belden.com read with fetch_source.py pdf and discovery by WebSearch; newest advisory PSIRT-6 of 2026-09-09
f5-k-articleshttps://my.f5.com/manage/s/article/K000163925bridge:extract → webfetch → bridge:jina200 js-shell
myF5 Salesforce single-page application: extract, WebFetch and the reader return an empty shell
F5 text read through the CNA record, nginx.org advisories, Rapid7, Wiz and CERT.at
baselland-chhttps://www.baselland.ch/politik-und-behorden/regierungsrat/medienmitteilungenbridge:extract → webfetch → bridge:url403 transport-403
canton site answers 403 on every transport
communiques read verbatim through the Polizeiticker relay and four regional outlets; the incident is held on the backlog
airlock-advisoryhttps://www.airlock.com/security-advisoriesbridge:extract → bridge:url-direct → webfetch400 js-shell
cookie-check 400 shell; WebFetch shows a disclosure policy with no advisory entries; Techzone needs a login
Airlock CNA text read through the ENISA EUVD record and the vendor documentation changelogs

Bridge invocations (this run)

7 bridge calls this run · these are successful bridge fetches (separate from "Coverage gaps" above).

7 other
  • 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 ncsc-csh post ×1
  • tools/fetch_source.py pdf ×1
  • WebFetch with the outbound-links template ×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 · 41 findings (truth=17, editorial=17, advisory=7) · Claude Sonnet 5.5 · 32m 57s

F-codeSectionItem · URL/quoteVerifier summaryRemediation · outcome
F3
claim-not-supported
·
The Obsidian page's own byline reads 'Published: June 11, 2026 | Updated: September 7, 2026' and its timeline ends 'June 11, 2026 - Published the research'; 2026-06-15 is The Hacker News's date. A 4-dfixed: the Obsidian source record and every inline citation of its page now carry 2026-06-11
F3
claim-not-supported
·
Adjacency: the Obsidian page carries the 9.9 chain score, master/salt key, DATABASE_URL and man-in-the-gateway text but no VulnCheck 8.8 (3.1) score (0 matches); that figure is on https://thehackernewfixed: The Hacker News is cited at the VulnCheck score clause
F3
claim-not-supported
·
The CERT-Bund CSAF for WID-SEC-W-2026-4427 carries aggregate_severity 'kritisch' (critical), and the bulletin's revision 2 (2026-11-16/17, summary 'Exploit aufgenommen') records that an exploit existsfixed: the CERT-Bund bulletin is stated as critical, with its second revision's exploit note
F3
claim-not-supported
·
Adjacency: Unit 42 gives the POST URL with __passthru=%2569ndex.php and route=DataImportController:index but says nothing of a php:// stream or the file-name field; that detail is Securifera's (POST[ffixed: Detection separates Unit 42's passthrough and route POST from Securifera's php:// stream, which needs body-capable logging
F3
claim-not-supported
·
The cited ENISA record says 'buffer overflow' and never 'heap' (0 matches); 'heap' and 'HTML entity decoder' come from the CVE record's CNA title 'Heap out-of-bounds write in the Smart Form Protectionfixed: 'heap' and 'entity decoder' removed from the title and summary; the cited record says buffer overflow
F3
claim-not-supported
·
VMRay says newer Zer0day Stealer versions show 'file-extension-based theft targeting Office documents and VPN configuration files'. It says nothing of VPN access brokerage (the sourcing_note wording ifixed: reworded to 'without mentioning VPN access brokerage'
F3
claim-not-supported
·
The CyberPower product page's metadata date is 2022-01-13 and it carries no dateline, so 2026-11-19 is the retrieval date; the Airlock docs pages (changelog 8.6.1, 8.5.3, Smart Form Protection, releasfixed in part: the Apache thread is dated 2026-11-01; undated vendor and documentation pages are disclosed in the sourcing notes as carrying the retrieval date
F3
claim-not-supported
·
The parenthetical sits in the clause attributed to CERT-EU, whose page scores all five CVSS 8.1 (3.1) and never mentions 9.2 or attack complexity; the 9.2/AC:H comes from F5's CNA record, NCSC.ch, NCSfixed: the summary attributes CVSS 4.0 9.2 to NCSC.ch, NCSC-NL and Rapid7, not to CERT-EU
F4
hallucinated-fact
·
The cited Tenable page (updated Nov 18) says 'As of November 18, GreyNoise has reported exploitation attempts against CVE-2026-114938, though GreyNoise has not confirmed any successful exploitation agfixed: title, headline, summary, body and sourcing note now report GreyNoise's exploitation attempts and VulnCheck's customer exploit; Breen's claim reworded to 'no release stated'
F4
hallucinated-fact
·
Mandiant limits rotation and web-root inspection to 'Environments with evidence of exploitation should rotate all secrets stored in Web.config and applicationHost.config ... and inspect webroots for ufixed: summary and action state Mandiant's condition for rotation and label the broader step
F4
hallucinated-fact
·
Symantec describes 'the typical pattern of the DLL side-loading technique' (upnpcont.exe loading atl.dll). In the pinned ATT&CK v20.1 T1574.001 is 'DLL' (search-order hijacking) and T1574.002 is 'DLL declined: the pinned ATT&CK v20.1 has T1574.002 revoked and T1574.001 ('DLL') as its survivor, so the mapping is the active id
F4
hallucinated-fact
·
Symantec says the product's own event-action feature runs administrator-chosen .cmd scripts via cmd /c under ppbed.exe ('looks like a power-management product launching cmd.exe'); its krbtgt advice apfixed: the action trigger is tied to the Triage discriminators (scripts differing from defaults, children that are not shutdown or notification routines, no preceding power event)
F4
hallucinated-fact
·
Neither the Obsidian page nor The Hacker News says this: Obsidian only says 'Upgrade to v1.83.14-stable or later. This is the first release that contains the complete fix set' (searches for under-deplfixed: the under-deployment claim is replaced by Obsidian's statement that v1.83.14-stable is the first release with the complete fix set
F4
hallucinated-fact
·
Cyderes says only that forti_vpn.go 'Maintains persistent connectivity with a 30-second reconnect loop' (a retry interval in the wrapper). A session that is sustained does not reconnect every 30 seconfixed: the 30-second interval is presented as the client's retry loop in Detection and the action
F4
hallucinated-fact
·
SA-CORE-2026-018 (listed in the same sentence's section) is a 'Gadget chain' in core's unserialize path, a flaw outside the installer; the sentence contradicts the section's first paragraph. If it meafixed: replaced by 'neither advisory reports exploitation'
F4
hallucinated-fact
·
The per-CVE authority (VulnCheck CNA record) says 'LiteLLM prior to 1.83.10' (CVE-2026-47101: prior to 1.83.14). The chain-level fix is v1.83.14, but the per-CVE field should follow the record, or thefixed: CVE-2026-47102 affected and fixed follow the VulnCheck record (1.83.10), with the chain fix labelled
F13
?
·
The changelog line 'FIX: AP-38633 Fix buffer overflow when decoding HTML entities' carries no CVE id, no mention of Smart Form Protection and no security label; no cited page states that it is the CVEfixed: reworded as a fix in the same releases, without a CVE id
F5
missing-citation
·
No cited page names 8u513-perf as a fixed build: Oracle's CSPU lists only 8u511b33-perf as affected and Upwind lists fixed .1 builds for 17, 21, 25 and 27 only. Oracle's 8u513-perf release notes (datedeclined: statement carried since the entry's first publication and confirmed by its earlier double-CLEAN; low confidence in the finding; Oracle's 8u513-perf release note supports it
F5
missing-citation
·
The cited portal page is a JavaScript shell on extract and WebFetch returned nothing; the statement is not in the CSAF JSON or the text advisory read this pass (both carry the version ranges, workaroufixed: the sentence is replaced by the CSAF's version scoping, cited to the CSAF
F9
surface-contradiction
·
Talos (2026-11-18) and the NCSC.ch hub say no malicious exploitation is established; Tenable (cited, updated Nov 18) relays GreyNoise exploitation attempts and scanning, and VulnCheck reports a commerfixed: GreyNoise's attempts and VulnCheck's customer exploit are stated beside Talos and NCSC.ch, and a contradiction line is in the sourcing note
F9
surface-contradiction
·
The CVE record (MITRE CNA) scores CVSS 3.1 8.8 AV:N/AC:L/PR:L (low privileges), while the vendor GHSA scores CVSS 4.0 5.9 Moderate and says a proxy-admin credential is needed. The entry follows the GHfixed: the score is labelled '5.9 (CVSS 4.0, vendor)' and the CNA's 8.8 conflict is in the sourcing note
F9
surface-contradiction
·
SA-24:20's revision table runs 2024-11-05 to 2024-11-14 (fixes for DSM 7.2.1 published 2024-11-12) and BleepingComputer calls it 'a November 2024 security advisory'; Synology's notice says December 20fixed: the sourcing note names the December 2024 versus November 2024 disagreement
F8
needs-more-research
·
The cited sources give interim measures for the rest: Xint 3.3.2 (fixed upstream group instead of request-time DNS lookups for CVE-2026-123634), 3.3.3 (remove QUIC listeners and http3 for CVE-2026-121fixed: interim measures for CVE-2026-123634, CVE-2026-121810 and CVE-2026-113425 added from Xint and CERT.at
F8
needs-more-research
·
Xint, the source of the name, chained CVE-2026-123007 with a separate unpatched Linux kernel SCTP flaw to 'gain root privileges on the host' and escape a Docker container ('Until a kernel patch is avafixed: Xint's SCTP chain to host root and container escape added with its mitigation
F8
needs-more-research
·
F5's own framing (CNA records, CERT.at 'Auswirkungen') is worker crash or memory corruption, with code execution 'on systems with ASLR disabled or when the attacker can bypass ASLR'; CERT-EU, Wiz, Rapfixed in part: Wiz's statement that ASLR does not protect is added; F5's ASLR prerequisite sits only in the CNA record, which cannot be cited
F8
needs-more-research
·
The update removed three source-supported hunting facts without saying so: Upwind's 'bursts of IllegalArgumentException from java.util.Base64$Decoder' and 'bursts of restarts', Equixly's 'Do not expecfixed in part: Upwind's exception and restart hooks, Equixly's caution and the Fortinet split are restored; Talos's Snort coverage is not, for the analysis word bound
F8
needs-more-research
·
Obsidian says the production routes 'POST /guardrails and PUT /guardrails/{id}' run user code with exec() and are 'tracked under CVE-2026-40217', so the vendor workaround of blocking only test_custom_fixed: the production routes POST /guardrails and PUT /guardrails/{id}, Obsidian's post-compromise audits and the endpoint block are in the entry and the action
F8
needs-more-research
·
Symantec's Recommendations give the levers ('restrict TCP port 3052 to named administrator hosts', 'Change the default administrator password, ensure patches are up to date') and CISA's ICSA-24-123-01fixed in part: hardening levers and the observed download and in-memory behaviours added, T1140 and T1620 mapped
F8
needs-more-research
·
Mandiant notes that if the <csrfSettings> section is omitted from Web.config, 'a setter bug in CsrfSettings.DetectionResult overwrites the fallback FormFieldName value with "RaiseException"', so the dfixed: the setter-bug caveat on the default field name added to Detection
F8
needs-more-research
·
The advisory's risk legend reads 'E - Exploit: Proof of concept exists (documented methods for developing exploit exist in the wild)' and 'TD: Only uncommon module configurations are exploitable'. Thefixed in part: glosses of E and TD added to the body and Exposure; status unchanged because the legend's wording is not a public proof of concept
F8
needs-more-research
·
SA-26:13's table lists fixes only for DSM 7.4, 7.3, 7.2.2 and 7.2.1; the SA-24:20 builds named for 7.2, 7.1 and 6.2 close CVE-2024-10441 but leave the September flaws (no mitigation, no fix) open on tfixed: the action says DSM 7.1, 7.2 and 6.2 owners also need a branch with an SA-26:13 fix
F10
missed-angle
·
A confirmed, ongoing incident at a Swiss canton (attacks since 2026-11-15, servers isolated from the Internet, about 20 applications on six servers being reinstalled, a precautionary Alertswiss noticedeclined: no source gives a vector, actor, data or behaviour, so an incident entry would carry no evidence-bound technique and check_run FAILs that; held with a named condition and expiry (the Beyond Gravity precedent)
F10
missed-angle
·
Section 6 of the Xint write-up describes a kernel SCTP flaw (association transferred to another socket, freed by the waiting send) chained to root and container escape, with details withheld until a kdeclined: details withheld and no CVE or patch exists; Xint's chain is now stated in the nginx entry
F16
?
·
An unauthenticated RCE exploited since 2026-10-07 (Unit 42), on the CISA KEV list since 2026-11-18, with a public PoC and a fix, on a product the store's own entry of 2026-06-16 (PRC UNC6508 via REDCafixed: priority high, the earlier REDCap entry linked in references and GTIG's UNC6508 reporting cited
F11
editorial-advisory
·
Style rule: zero IOCs including file-name indicators. The implant and side-loaded DLL names and the ransomware extension and note name are family-specific file-name indicators; describe the behaviour fixed in part: implant and DLL names removed from the CyberPower entry; the Synology entry's extension and note names are published by the vendor and 360 and stay (advisory)
F11
editorial-advisory
·
The KEV window tool states for this row: 'the KEV listing is an exploitation-status change (PD-13), an `update` record on that entry'; the run used `improvement` (no re-float) while the same listing ofixed: the record is retyped update, its section heading and updated_at moved with it
F11
editorial-advisory
·
The relative clause attaches the clipper and skimmer to the 450 domains; say the Optimizely files carry a clipper and skimmer.fixed: summary clause reworded
F11
editorial-advisory
·
Record-keeping narration in reader-facing text. Obsidian (cited) still describes the production guardrail routes as running user code with exec() and builtins available, so the 'correction' also narrofixed: record-keeping narration removed from the section
F11
editorial-advisory
·
The run registered product:redcap, product:airlock-gateway, product:idunno-anticsrf and product:cyberpower-powerpanel-business, and the registry already holds product:f5-nginx-open-source and product:fixed: product entity keys added to the new entries; the unsourced Swiss region dropped from the Airlock entry
F11
editorial-advisory
·
Fetch-transport vocabulary in a reader-facing field; two sentences of provenance suffice ('F5's K-articles are not publicly readable; Plus builds are taken from CERT.at, Rapid7 and Wiz').fixed: 'not publicly readable' replaces the transport wording
F11
editorial-advisory
·
The body relays Cyderes's reading of 'brute-force harvesting and credential dump sourcing' and the broker's brute-force claim but no Brute Force id is mapped (T1110); either map it as the reported behdeclined: both statements are attributed to Cyderes and the broker and neither is an observed behaviour, so no T1110 is mapped

Iteration #2 NEEDS_FIXES · 22 findings (truth=14, editorial=4, advisory=4) · Claude Sonnet 5.5 · 23m 51s

F-codeSectionItem · URL/quoteVerifier summaryRemediation · outcome
F3
claim-not-supported
·
The alert lists the row as 'CVE-2026-90436 Apache Axis Unsafe Reflection Vulnerability', so it does name a product (Apache Axis); the record summary repeats it as '(no product named)'. Reword to 'namifixed: the section and record summary now say the alert names neither Confluence nor N-able N-central
F3
claim-not-supported
·
Obsidian's rotation advice reads 'Rotate credentials. Rotate provider API keys, database credentials, and any MCP tokens stored in the proxy if exposure is suspected.' It names neither the master key fixed: the rotation list follows Obsidian (provider API keys, database credentials, MCP tokens); the master and salt keys are stated as the entry's addition
F3
claim-not-supported
·
CERT.at says '37.2.0.1 und 37.0.7.1 enthalten bereits einen Fix fuer CVE-2026-123633, sind jedoch weiterhin von CVE-2026-123007 und CVE-2026-121810 betroffen' and does not say 'only'; its own affectedfixed: the sentence states what CERT.at says about 37.2.0.1 and 37.0.7.1
F3
claim-not-supported
·
'Them' is the five flaws. NCSC.ch's post scores only CVE-2026-123633 and CVE-2026-123007 at 9.2, Rapid7 only CVE-2026-123007; only NCSC-NL lists all five at CVSS (v4) 9.2. Say 'NCSC-NL scores all fivefixed: the summary attributes the scores to NCSC-NL (all five) and to NCSC.ch and Rapid7 (the HTTP/2 pair)
F3
claim-not-supported
·
The PSIRT page lists 22 further products under 'Products confirmed NOT Impacted' (FortiRecon, FortiClient Windows/Linux/macOS, FortiAuthenticator agents, FortiToken, FortiNAC and others), so 'the othefixed: the Fortinet clause says dozens under investigation and others not impacted
F3
claim-not-supported
·
The hub says 'Identify exposure: Record the Java distribution, exact running build, processor capabilities and responsible supplier.' and says nothing about an installed package versus a running buildfixed: the running-version point is cited to BACS
F3
claim-not-supported
·
The citations read 'Airlock documentation, 2026-11-18', which is the CVE publication date, while the pages were retrieved on 2026-11-19 (the CyberPower entry's vendor page correctly carries 2026-11-19fixed: the documentation citations and source records carry 2026-11-19, the retrieval date
F4
hallucinated-fact
·
Residual of iteration 1 (title and summary were corrected, the headline was not). The cited ENISA record says 'The buffer overflow then happens during the parsing of that HTML response' and never 'heafixed: headline now says buffer overflow
F4
hallucinated-fact
·
Rapid7's remediation table says 'NGINX Plus | R34 - R36 | None' and CERT-EU says 'The versions R34, R35 and R36 are also affected'; CERT.at's affected list reads 'R34 bis R36 (alle vier Schwachstellenfixed: per-CVE NGINX Plus R-release ranges follow CERT.at, Rapid7 and CERT-EU (R34 to R36; R33 only for CVE-2026-113425)
F4
hallucinated-fact
·
Residual of iteration 1. Symantec ties the krbtgt advice to finding the intrusion's techniques and indicators ('Review the techniques and indicators in this document. If you find them, assume Active Dfixed: the action no longer resets krbtgt on a script that differs from defaults; it follows Symantec's trigger (its listed techniques) and its advice to treat unknown scripts as hostile
F4
hallucinated-fact
·
Symantec says the implant name 'is a name associated with a Microsoft UPnP component' and calls the pairing 'an old, trusted executable paired with an unknown DLL in a user-profile path ... the typicafixed: 'signed-looking' replaced by Symantec's wording (old trusted executable, unknown DLL in a user-profile path)
F4
hallucinated-fact
·
The SA-24:20 builds close three flaws (CVE-2024-10441, CVE-2024-50629 and CVE-2024-10445), not only CVE-2024-10441. The point the action wants (they leave SA-26:13's flaws open because that bulletin lfixed: 'only the SA-24:20 flaws'
F4
hallucinated-fact
·
The CVE record's cveMetadata reads assignerShortName 'mitre' (CNA MITRE, published 2026-04-10) with CVSS 3.1 8.8 AV:N/AC:L/PR:L; VulnCheck is the CNA for CVE-2026-47101 and CVE-2026-47102 only. Say 'tfixed: 'the CVE record published by MITRE'
F5
missing-citation
·
Neither the Securifera post nor the OpenCVE page cited states that Securifera is the CNA (the CVE record's assignerShortName is 'Securifera', but that record is not cited). Cite a page that carries itfixed: 'and CNA' dropped from the Securifera descriptions
F5
missing-citation
·
Mandiant limits rotation to 'Environments with evidence of exploitation' and says to review IIS logs for POSTs with elevated cs-bytes 'where logged'; 'request bodies are rarely logged' is in no cited fixed: the broader rotation is labelled a precaution beyond Mandiant's condition and the unsourced premise is dropped
F8
needs-more-research
·
The advisory data fetched this pass carries revision 13.0 of 2026-11-18: 'ONTAP 9.15.1P21 and ONTAP 9.18.1P9 added to Software Versions and Fixes' (revision history 10.0 to 13.0). The entry, updated ofixed: ONTAP 9.15.1P21 and 9.18.1P9 added to fixed, the body, the sources (revision 13.0) and the section
F8
needs-more-research
·
Residual of iteration 1: the remediation says F5's ASLR prerequisite 'sits only in the CNA record, which cannot be cited', but CERT.at (cited, read this pass) states it: 'Auf Systemen, auf denen Addrefixed: CERT.at's statement of F5's ASLR condition and base effect added; headline reworded to 'can reach worker RCE'
F11
editorial-advisory
·
The exploitation-status change (GreyNoise exploitation attempts as of 2026-11-18, VulnCheck's customer exploit) is the new fact behind the changed title, headline and summary, but neither the update sfixed: the section and record summary carry GreyNoise's attempts and VulnCheck's customer exploit
F11
editorial-advisory
·
Iteration 1 asked for file-name indicators to go; the implant and DLL names were removed, but the attacker script names (the pp-plus-five-letters pattern, ifm.cmd to ifm4.cmd) remain. Describe the behfixed: attacker script names removed; behaviour described as scripts staged under extcmd whatever their name (the Synology extension and note names stay as vendor-published, advisory)
F11
editorial-advisory
·
The run registered these product keys, but the CyberPower entry carries only the incident key, the Java4Shell entry has entities: [] while its affected_products name the three Cisco products, and the fixed: product keys added to the entities of the CyberPower, Java4Shell and PHP2Shell entries
F11
editorial-advisory
·
Iteration 1 moved the Obsidian primary to its own date (2026-06-11, 'June 11, 2026 - Published the research'); event_date still carries The Hacker News's 2026-06-15. Align event_date with the primary fixed: event_date follows the Obsidian primary, 2026-06-11
F13
?
·
Cyderes says the credential file 'Contains a mix of default credentials (guest:guest, admin:admin, test:test) and organization-specific passwords' and that logs confirm tunnels into multiple victims; fixed: the headline names the credential file's mix and the two VPN clients, and the title the toolkit's VPN clients

Iteration #3 NEEDS_FIXES · 17 findings (truth=7, editorial=8, advisory=2) · Claude Sonnet 5.5 · 25m 18s

F-codeSectionItem · URL/quoteVerifier summaryRemediation · outcome
F4
hallucinated-fact
·
Residual of iteration 2 F4 #3: only actions[0] was changed. Symantec ties the reset to its own techniques and indicators ('Review the techniques and indicators in this document. If you find them, assufixed: the takeaway reset is tied to the techniques Symantec lists, the Triage discriminators drop 'differs from product defaults' for unconfigured scripts and assignments, and the entry-derived filter is removed from…
F3
claim-not-supported
·
Residual of iteration 2 F3 #1 (fixed in the section and record summary, not in the sourcing_note). The CISA alert lists the row as 'CVE-2026-90436 Apache Axis Unsafe Reflection Vulnerability', so it dfixed: the sourcing note says naming neither Confluence nor N-able N-central and no actor
F13
?
·
(low confidence) Residual of iteration 2 F13. The remediation record says the headline 'names the credential file's mix and the two VPN clients', but the headline still names only 'default logins' as fixed: the headline names the credential file's mix of default and real logins
F3
claim-not-supported
·
(low confidence) The page never says Symantec could not establish the route. It says 'Any of these, or stolen administrator credentials, could give an attacker the necessary console access to carry oufixed: 'Symantec does not say how the attacker reached the console' in the body, summary and title
F3
claim-not-supported
·
(low confidence) Tenable: 'GreyNoise has reported exploitation attempts against CVE-2026-114938, though GreyNoise has not confirmed any successful exploitation against its sensors.' The entry drops 'afixed: 'none confirmed successful against its sensors' everywhere it appears
F3
claim-not-supported
·
(low confidence) Obsidian's advice is 'Rotate credentials. Rotate provider API keys, database credentials, and any MCP tokens stored in the proxy if exposure is suspected.' The entry keeps Obsidian's fixed: Obsidian's condition (where exposure is suspected) leads the rotation advice, and the broader trigger and the master and salt keys are stated as the entry's addition
F3
claim-not-supported
·
(low confidence) The clause is cited to CERT.at alone, which says the Plus builds 'beheben alle vier Schwachstellen' and lists CVE-2026-123007, -121810, -123633 and -113425; CVE-2026-123634 is not in fixed: the sentence states what CERT.at lists (four of the five flaws) and the sourcing note says the CVE-2026-123634 Plus builds are presumed
F9
surface-contradiction
·
The R33 variance is attributed to CERT.at, but CERT.at lists 'R34 bis R36 (alle vier Schwachstellen)' and R33 only for CVE-2026-113425. The cited source that gives R33 for the HTTP/2 flaws is Wiz: tabfixed: the sourcing note names Wiz's R33 and states the majority reading
F8
needs-more-research
·
CERT.at: 'NGINX Plus Rx: R34 bis R36 (alle vier Schwachstellen)' and 'Für NGINX Plus R33 bis R36 sind derzeit keine Updates verfügbar'. CVE-2026-121810 is one of the four, yet its affected and fixed ffixed: the CVE-2026-121810 record carries R34 to R36 and none-for-R34-to-R36
F5
missing-citation
·
(low confidence) No cited page calls Airlock a web application firewall or API gateway vendor or the product a WAF (ENISA carries 'Airlock by Ergon', assigner Airlock, published Nov 18; the documentatfixed: 'web application firewall and API gateway vendor' and 'WAF' replaced by the sources' wording
F5
missing-citation
·
(low confidence) True when checked (the CVE record's CISA-ADP SSVC block reads Exploitation none, timestamp 2026-11-19T01:39:42Z, ten minutes after the 01:29Z publication), but the CVE record is not afixed: the CISA-assessment clause dropped from the sourcing note (the CVE record is not a listed source)
F5
missing-citation
·
(low confidence) The line carries no citation and no 'inference' label (the Confluence entry labels its equivalent). The endpoints and the user_role escalation are Obsidian's, the process-lineage huntfixed: the Detection line cites Obsidian for the endpoints and the user_role escalation and says the process-lineage hunt follows from the code execution
F8
needs-more-research
·
(low confidence) The advisory (updated 2026-11-18) lists Deep Discovery Analyzer as 'Impacted' and gives 'Engine: Fixed via Active Update on Nov.6. OS Image: 7.7: Critical Patch 1020 7.6: Critical Patfixed: the Analyzer OS-image scope and the Active Update engine fix (2026-11-06) are in cves[].affected
F16
?
·
(low confidence) The update moved the entry from notable to high. The entry's own text says Drupal rates the one critical flaw as exploitable only in uncommon configurations (installer not secured or fixed: priority returned to notable (doubt about a non-default-configuration flaw with no reported exploitation resolves to notable) and the record summary no longer states a priority move
F18
?
·
(low confidence) The action chains four tasks, restates the Detection IIS-log hunt, and ends in a hedged clause ('a candidate for the same rotation as a precaution') that is not executable without re-fixed: the action keeps the search-and-remove task and the evidence-conditional rotation; the precaution clause is dropped
F11
editorial-advisory
·
Record-keeping narration about the entry's own earlier text in reader-facing prose (section and record summary). The exploitation-status change itself is stated; the clause about what the entry used tfixed: the narration clause removed from the section and the record summary
F11
editorial-advisory
·
(low confidence) The Unit 42 note is dated 2026-10-07 and says 'Attackers are actively exploiting CVE-2026-90817'; it gives no start date, so 'from 2026-10-07' reads as a start date ('as of 2026-10-07fixed: 'exploited as of 2026-10-07'; the sourcing note says the event date is the CISA listing and gives the disclosure date

Iteration #4 NEEDS_FIXES · 16 findings (truth=8, editorial=5, advisory=3) · Claude Sonnet 5.5 · 23m 34s

F-codeSectionItem · URL/quoteVerifier summaryRemediation · outcome
F3
claim-not-supported
·
Rapid7's page scores only CVE-2026-123007 (table row '| CVE-2026-123007 | HTTP/2 flow-control memory corruption permitting worker-context RCE | CWE-191 | 9.2 (Critical)'), not the HTTP/2 pair. NCSC.chfixed: the summary now says NCSC-NL scores all five, NCSC.ch and Wiz the two HTTP/2 flaws and Rapid7 CVE-2026-123007
F4
hallucinated-fact
·
No source says the Plus builds carry all five fixes. CERT.at: 'F5 stellt Updates bereit, die alle vier Schwachstellen beheben' (CVE-2026-123007, -121810, -123633, -113425; CVE-2026-123634 absent). Thefixed: the CVE-2026-123633 record says per CERT.at the Plus builds carry the fixes for four of the five
F3
claim-not-supported
·
(low confidence) nginx.org lists more than five advisories, each with a severity and version range (e.g. 'Use-after-free in HTTP/3 Severity: medium'); it says nothing about authentication and does notfixed: nginx.org now carries only 'five memory-corruption flaws in the data plane'; 'unauthenticated' sits on the CERT-EU clause that states it
F4
hallucinated-fact
·
(low confidence) Wiz states 'we will not be releasing these exploits at this time'. The Xint page makes no statement about its nginx exploits; it says only 'Until a kernel patch is available, we are wfixed: the sourcing note says Wiz did not publish its exploits and Xint reports code execution for the flaws it found
F3
claim-not-supported
·
The Securifera post is dated 2026-09-27 and states no disclosure date. 2026-09-20 is the CNA record's publication date (OpenCVE: 'Published Sep 20, 2026, 12:05 PM', CNA Vanderbilt University; Pentest-fixed: the summary says Securifera described the flaw on 2026-09-27
F3
claim-not-supported
·
The sentence ends in 'as Obsidian advises' and one Obsidian citation. Obsidian's self-host list is: upgrade, audit user roles, audit guardrails, audit callbacks, rotate credentials if exposure is suspfixed: the LiteLLM-sourced upgrade and endpoint block are cited to the vendor advisory; Obsidian's advice (suspected exposure, rotation, audits) is a separate sentence; the admin-endpoint clause is stated as the entry's…
F4
hallucinated-fact
·
The entry's frontmatter reads 'priority: notable' after iteration 3 returned it from high, and its record summary states no priority move. The note still says notable to high.fixed: the run record notes say priority stays notable
F4
hallucinated-fact
·
The Confluence entry's last record is typed `update` (at 2026-11-19T05:05:36Z) and updated_at moved to it; iteration 1 retyped the record (run record finding: 'the record is retyped update, its sectiofixed: the run record notes say the Confluence record is typed update and moves updated_at
F9
surface-contradiction
·
(low confidence) Mandiant's timeline reads '2026-11-18: Mandiant advisory and CVE-2026-131914 published', while the ENISA record the entry cites gives the CVE 'datePublished Nov 19, 2026, 1:29:12 AM' fixed: the sourcing note gives both publication dates (Mandiant 2026-11-18, ENISA record 2026-11-19)
F5
missing-citation
·
(low confidence) True when checked (CVE record cvssV3_1 8.8, PR:L, published 2026-04-10) but the CVE record is not a listed source and no listed page carries it. Iteration 3 dropped the same kind of ufixed: the MITRE-record clause is dropped from the sourcing note; the recorded figures are the vendor's
F5
missing-citation
·
(low confidence) The hunt is the entry's own inference from the install.php mechanism, with no citation and no inference label (other entries label theirs). NCSC-NL ('when the installer is not correctfixed: the Drupal Detection line says the hunt follows from the mechanism
F8
needs-more-research
·
(low confidence) Obsidian names /key/generate with an explicit allowed_routes array as 'the cleanest path' of the first step (CVE-2026-47101; same unchecked write on /key/update, /key/regenerate, /keyfixed: Detection adds /key/generate calls carrying allowed_routes and /user/bulk_update
F18
?
·
(low confidence) Restates the Detection line's Zscaler guidance (review account-creation events and administrative access, audit all Project JavaScript fields, rotate API tokens, session tokens and infixed: the action no longer restates the Detection line
F11
editorial-advisory
·
Stray comma before the citation parenthesis in the update section (the main-analysis copy of the clause has none).fixed: the stray comma removed
F11
editorial-advisory
·
(low confidence) The JSON and TXT URLs appear once with the 2026-10-22 revision-1 date and again with the 2026-11-17 revision-2 date, both primary, and both URLs now serve revision 2. Duplicate links declined: advisory; the repeated URL records carry the revision-1 and revision-2 dates of one bulletin, the same pattern the entry uses for each revised vendor page
F11
editorial-advisory
·
check_run.py WARN citation-cve: the clause names CVE-2026-123634 but its CERT.at citation does not mention that id (CERT.at lists CVE-2026-123007, -121810, -123633 and -113425). The zero-warning rule fixed: the sentence names the four CVE ids CERT.at lists

Iteration #5 NEEDS_FIXES · 5 findings (truth=3, editorial=1, advisory=1) · Claude Sonnet 5.5 · 17m 34s

F-codeSectionItem · URL/quoteVerifier summaryRemediation · outcome
F3
claim-not-supported
·
Introduced by the iteration-4 remediation. The advisory the clause cites says 'Fixed in 1.83.11 ... Upgrade to 1.83.11 or later.' and 'block POST /guardrails/test_custom_code at your reverse proxy or fixed: v1.83.14-stable is cited to Obsidian and the endpoint block to the LiteLLM advisory
F14
?
·
(low confidence) Left over from iteration 4: the remediation narrowed the clause but kept 'five' on nginx.org. nginx.org lists ten advisories fixed in 1.31.9/1.30.7 (CVE-2026-123634, -123007, -121810,fixed: the count of five is attributed to CERT-EU; nginx.org is cited for the publication of the fixes
F4
hallucinated-fact
·
(low confidence) After iteration 4 dropped the CVE-record clause, no listed source gives a score for CVE-2026-47101: The Hacker News relays only 'VulnCheck, which assigned the CVE, scores it 8.7 underfixed: CVE-2026-47101 carries cvss n/a and the sourcing note says no listed source scores it on its own
F9
surface-contradiction
·
(low confidence) CERT.at, a listed source, scopes this CVE to 'CVE-2026-113425: 37.1.0 bis 37.2.0 sowie 37.0.0 bis 37.0.7', which leaves 37.2.0.1 and 37.0.7.1 outside it. F5's CNA record for CVE-2026-fixed: the CVE-2026-113425 Plus range follows CERT.at, a listed source
F11
editorial-advisory
·
(low confidence) The iteration-4 reword scoped the action to Optimizely account holders and added the cause, but the audit and rotation steps are still the Zscaler guidance the Detection line carries fixed: the Detection tail is shortened to the account-history review; the action keeps the audit and rotation

Iteration #6 CLEAN · 5 findings (truth=0, editorial=0, advisory=5) · Claude Sonnet 5.5 · 18m 15s

F-codeSectionItem · URL/quoteVerifier summaryRemediation · outcome
F11
editorial-advisory
·
(low confidence) The record summary ends in a clause about what the entry used to say. The update section carries no such clause, so summary and section differ, and the correction narration adds nothifixed: the record summary no longer narrates what the entry used to say
F11
editorial-advisory
·
(low confidence) The score carries no scale although the body says 8.8 (3.1) and CVE-2026-40217 is labelled (CVSS 4.0, vendor); 'the VulnCheck record' is not among sources[] while Obsidian's timeline,fixed in part: the CVE-2026-47102 score now carries its scale and source (CVSS 3.1, VulnCheck); the relayed-by wording on the affected and fixed ranges stays because the version comes from the VulnCheck record as The…
F11
editorial-advisory
·
(low confidence) Obsidian: 'Any user who can reach these handlers can promote themselves to proxy_admin: an org_admin can do it directly, and a default internal_user can do it after CVE-2026-47101.' Tleft: advisory at low confidence; the body states the guardrail endpoint needs a proxy-admin credential by default and that the chain's escalation supplies it, which is how Obsidian and LiteLLM describe it
F11
editorial-advisory
·
(low confidence) The page is a living listing with no dateline (extract metadata reads 2010-01-13; no text dates the fixes); 2026-11-18 is the release date taken from CERT-EU, Rapid7 and Wiz. The sourfixed: the sourcing note says the nginx.org advisories page has no dateline and the date comes from the other sources
F11
editorial-advisory
·
(low confidence) git diff HEAD shows 'priority: notable' as an unchanged line, so the record names a field that did not move (priority); updated_at follows from the record type and the other update releft: advisory at low confidence; the record declares priority because the entry's priority was revised back to notable, which the earlier working-tree state had moved

Iteration #7 NEEDS_FIXES · 8 findings (truth=3, editorial=1, advisory=4) · Claude Sonnet 5.5 · 28m 48s

F-codeSectionItem · URL/quoteVerifier summaryRemediation · outcome
F3
claim-not-supported
·
Equixly's page (dated 2026-11-17) does not expect a crash: its 'What is the impact?' section says 'The worst case is code execution as the user running the JVM' and its TL;DR says 'Calif reports remotfixed: the sourcing note no longer characterises Equixly; the Calif demonstrations clause stays
F3
claim-not-supported
·
The SA-24:20 revision table runs to 2025-03-27, not 2024-11-14: rev 4 2024-11-26 (DSM 7.1 fix), rev 5 and 6 2024-12-05 (DSM 6.2 added and fixed), rev 7 to 9 2025-03-19/20/27. December 2024 revisions tfixed: the sourcing note now cites SA-24:20's revision history (first published 2024-11-05, fixes on 2024-11-26 and 2024-12-05), which fits Synology's December 2024
F3
claim-not-supported
·
(low confidence) The live page text carries later dates: 'counted on 18 November 2026' and 'Latest match | 18 November' (and its fetch metadata reads 2026-11-19). 2026-11-17 is not the latest date itsfixed: the sourcing note gives the citation date as nominal and says the page's text counts through 2026-11-18
F8
needs-more-research
·
(low confidence) Tenable, cited four times, says 'As of November 18, Red Hat and Ubuntu had not released fixes, and Debian had fixed only its unstable branch' and that on 2026-11-18 'the Keycloak projfixed: the update section and the record summary carry Tenable's distribution fix status as of 2026-11-18 and the Keycloak interim mitigation
F11
editorial-advisory
·
(low confidence) Record-keeping narration in a reader-facing record summary, and the clause is not in the update section, so summary and section differ (check 4c(d)). Drop the trailing clause.fixed: the record summary drops the record-keeping clause
F11
editorial-advisory
·
(low confidence) The affected range and the target build are chained with the same word, so the task line reads as a three-step path. Write 'Update every Airlock Gateway on 8.5.0 to 8.5.2 to 8.5.3, anfixed: the action separates the affected range (8.5.0 through 8.5.2, 8.6.0) from the target builds
F11
editorial-advisory
·
(low confidence) All four ids are active in the pinned dataset. Each behaviour is described in the entry body and supported by a cited source but carries no id; the mapping is the canonical surface fofixed in part: T1068 and T1611 added to nginx, T1190 to OptimJack, T1069.002 to tcherber, T1027.003 to REDCap; T1046 left off tcherber because Cyderes says the port scanner is not called by the orchestrator and no use…
F11
editorial-advisory
·
(low confidence) The Unit 42 note is dated 2026-10-07 and says 'Attackers are actively exploiting CVE-2026-90817'; it states no start date. The entry itself says 'as of 2026-10-07' after an earlier itfixed: the run-record note says exploitation was reported as of 2026-10-07

Iteration #8 CLEAN · 3 findings (truth=0, editorial=0, advisory=3) · Claude Sonnet 5.5 · 22m 33s

F-codeSectionItem · URL/quoteVerifier summaryRemediation · outcome
F11
editorial-advisory
·
(low confidence) The clause follows 'Debian had fixed only its unstable branch'; Tenable says 'As of November 18, Red Hat and Ubuntu had not released fixes, and Debian had fixed only its unstable branfixed: the update sentence now says those distributions had no stable vendor build yet
F11
editorial-advisory
·
(low confidence) The entry covers only the VPN access-brokerage toolkit. The same Cyderes report also describes the operator's Rust loader, stealer, HVNC implant, Enigma encryptor and a C2 panel showileft: advisory scope choice; the entry covers the VPN toolkit that Cyderes' report centres on, and the loader, stealer and C2 panel add no behaviour a responder needs for this exposure
F11
editorial-advisory
·
(low confidence) The CERT-EU page says 'with a CVSS score of 8.1' and never names the scale (3.1 comes from CERT.at and the CNA record), and Rapid7's remediation table covers CVE-2026-123007 only ('NGfixed: the sourcing note drops the scale CERT-EU does not name and says Rapid7 gives the R34 to R36 range for CVE-2026-123007 only

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

2026-11-19T0355Z-intel · Sonnet 5.5 · window 26 h · 6 entries published

Verification & coverage notes

Window 26 h (gap 24.0 h to 2026-11-18T0356Z-intel). Six new entries and eight changelog records, no deep dive, no critical. The will-publish set was 14 items, over the ~8 a main-agent deep read should carry, so two scoped cti-research follow-ups read the primaries and returned claim tables whose passages were grep -F checked against saved bodies: DR1 (the six new entries, 2026-11-19T04:39:20Z to 05:05:15Z) and DR2 (the eight records, 04:39:21Z to 04:58:45Z); the main agent composed every entry and record from those tables and the findings files. Stop-hook commit nudges arrived mid-run and were not acted on; the commit follows the gate and the verifier loop.

Coverage and dispositions

  • KEV sweep (work/2026-11-19T0355Z-intel/kev-window.txt): CVE-2026-90817 REDCap → new entry 2026-11-19/redcap-cve-2026-90817-unauth-survey-rce-exploited-kev (Unit 42 had reported exploitation as of 2026-10-07, so the CISA listing was not the only evidence S1 reported); CVE-2024-10441 → update on the Synology DSM entry; CVE-2026-40217 → update on the LiteLLM entry (a v2-shaped legacy record, brought to the current schema in the same record: classification, techniques, affected_products, per-CVE fixed versions; the CVE's CVSS now follows the vendor advisory, 5.9, not an unsourced 8.8); CVE-2026-90436 → update on the Confluence entry (CISA's listing and Apache's no-fix and handler statements are the reader-facing delta; an earlier draft typed it improvement and the verifier asked for update, so it moves updated_at). No RANSOMWARE rows.
  • Backlog: no open rows at the start. Two rows opened and held with named conditions: the PHP critical release pre-announced for 2026-11-19 14:00 to 14:30 UTC (expiry 2026-11-26) and the Basel-Landschaft cantonal-administration incident (no vector, actor or behaviour, so no evidence-bound technique; expiry 2026-12-03).
  • Scheduled releases: Drupal core released 2026-11-18 (update on the pre-announcement entry; priority stays notable); NGINX and F5 2026-11-18 (new entry; BIG-IP rows dropped below); Oracle CSPU and TYPO3 carried by yesterday's entries with no revision; Microsoft revisions informational; Trend Micro KA-0024802 status of 2026-11-18 (update; DR2 corrected S1: the Mobile Security for Enterprise patch is on the Download Center dated 2026-11-17 although the advisory row still shows it pending).
  • Revised advisories: HPE HPESBNW05162 revision 2 (update; the entry carried CVE-2026-79845, which revision 2 replaced with CVE-2026-95787, and the FIPS build is now in the fixed strings); Java4Shell vendor impact from Cisco and Fortinet plus the BACS warning page (update; the entry cited BACS's 2026-11-17 announcement page for a compromise-check quote that only the 2026-11-18 warning page carries, so the citation was moved); OptimJack (update; Optimizely's status entry now names the flaw).
  • Source-slice rotation: 102 sources were excluded as attempted by the previous two fires; the breach pool (S4) is small, so nine of its ten records had to be drawn from recently attempted ones, oldest cursor first.
  • Corrections to the sub-agents' findings applied at composition: S2's claim that NGINX Plus 37.2.0.1 and 37.0.7.1 also fix CVE-2026-113425 was wrong (F5's record: only CVE-2026-123633); tcherber's orchestrator automates only the VPN connect, route parsing and LDAP enumeration, SMB reconnaissance and port scanning are manual tools; S1's list of Fortinet products as affected by Java4Shell was dropped (all are under investigation); CERT-EU states no exploitation for the nginx2root flaws, NCSC-NL only predicts it.
  • Single-source entries: Airlock Gateway (vendor CNA text via ENISA EUVD, corroborated by the vendor's release notes only), Idunno.AntiCsrf (Mandiant as CNA, sole observer of the exploitation; the CISA assessment entry in the CVE record lists exploitation as none), CyberPower PowerPanel (Symantec), tcherber (Cyderes; VMRay corroborates only the developer identity).
  • Candidate sources: nginx-security-advisories, unit42-timely-threat-intel, polizeiticker-ch (added); mitel-security-advisories promoted. Two source_health UNSOLVED items (typo3-security-advisories, cnil-fr) got health_cmd fixes that were not re-probed because the tool has no single-source mode; the next audit's run confirms them.
  • Operator signals: lab52 has failed on every transport for six fires (newest post 2026-10-05); the jina reader pool is dead (HTTP 402); cisa csaf-recent returned only a legacy bulk re-publish.

Borderline drops

  • borderline-drop: F5 BIG-IP K000164058 data-plane flaws, F5 BIG-IP K000164058: six data-plane flaws need non-default iRule commands or APM agents, no exploitation, F5 page unreadable
  • borderline-drop: Ruby 4.0.10, 3.4.13 and 3.3.14 memory-safety fixes, Ruby regex/case-conversion memory-safety fixes: CVE ids pending, no exploitation, needs attacker-controlled regex or crafted strings; patch-cycle
  • borderline-drop: Linux kernel NoisyTCP CVE-2026-99288, NoisyTCP CVE-2026-99288: reliable effect is remote panic only, UAF path hard, discoverer exploit lab-specific; kernel patch cycle
  • borderline-drop: Dell PowerStore DSA-2026-479, Dell PowerStore DSA-2026-479: internal storage service rows, no exploitation, no stated footprint
  • borderline-drop: Roundcube 1.6.21 and 1.7.6, Roundcube 1.6.21/1.7.6: no CVE ids, no exploitation, patch cycle
  • borderline-drop: Eclipse Mosquitto CVE-2026-110486, Mosquitto CVE-2026-110486: websockets listener only, no exploitation, niche
  • borderline-drop: Chrome 158 stable, Chrome 158 stable: 204 fixes, no exploitation statement, regular cycle
  • borderline-drop: Juniper JSA119568 Junos OS Evolved kernel TCP, Juniper JSA119568 Junos OS Evolved kernel TCP: vendor impact on an already-covered flaw, fix announced for later, no new action
  • borderline-drop: Broadcom AutoSys CA20261117-01, Broadcom AutoSys CA20261117-01: internal scheduler console, no exploitation, no stated footprint
  • borderline-drop: EU sanctions on NoName057(16), EU sanctions NoName057(16): no defender-side decision change
  • borderline-drop: Canton Bern ICSG in force 2026-11-01, Canton Bern ICSG in force 2026-11-01: sources dated 2026-09/10, outside window, awareness only
  • borderline-drop: UAT-12643 (Talos), UAT-12643 Talos: low-confidence DPRK attribution, Europe only potentially, no constituency nexus
  • borderline-drop: AuditTeam and OpsAgent (Securelist), AuditTeam/OpsAgent Securelist: single source, signed-driver killer and GitHub C2 are known classes; Italy/Finland victims only
  • borderline-drop: Check Point State of Ransomware Q3 2026; Check Point State of Ransomware Q3 2026: leak-site statistics report, stats not defender decision; feed stamp outside window
  • borderline-drop: Elastic analysis of the agent-framework compromise, Elastic agent-framework analysis: third analysis, no material delta on the covered entry
  • borderline-drop: Docker Model Runner (ZDI), ZDI Docker Model Runner: non-default configuration, single source
  • borderline-drop: ROMULUS and CNCMachineRMS (Blackpoint), Blackpoint ROMULUS/CNCMachineRMS: family naming, no victim or post-exploitation
  • borderline-drop: Equinor cyberattack, Equinor: no vector/actor, no constituency nexus
  • borderline-drop: Wellington IT credit-union platform, Wellington IT credit-union platform: out-of-nexus, single relay source
  • borderline-drop: TfL Oyster photocard site, TfL Oyster photocard site: out-of-nexus UK consumer accounts, single source
  • borderline-drop: Municipality of Neuried, Neuried municipality IT offline: no vector, no actor, below incident floor
  • borderline-drop: TeamCity CVE-2026-132231 to -132234, four CVSS 9.9 rows that need a low-privilege account, no exploitation (S1 drop list)
  • borderline-drop: Apache Roller CVE-2026-129448, PostgreSQL 14 to 18 minor releases, Apache Shiro and APISIX rows, SmarterMail CVE-2026-126969, Zyxel DX3301, Circutor PowerStudio, ASUS Control Center Express, DrayTek Vigor2960 and other EUVD 9.0+ rows, no exploitation and no stated constituency footprint (S1 drop list)
  • borderline-drop: Bitdefender EggStreme, Dream Mustang Panda ASEAN lures, ThreatFabric Bazuka, Natto ZRON report, SpecterOps Black October, out-of-nexus, paywalled or vendor commentary (S3 drop list)

Coverage gaps: lab52 (every transport fails, newest post 2026-10-05); belden-psirt (listing 403, advisory PDFs read); my.f5.com K-articles (single-page application, read through the CNA record and nginx.org); baselland.ch (403 on every transport, read through the Polizeiticker relay); airlock.com advisory page (disclosure policy only); consilium.europa.eu (403); cnil-fr (503 maintenance page); safeonweb-be (5xx outage ended); apa-ots, cyberattaque-org (retirement or horizon limits); dell.com, support.hpe.com, advisories.ncsc.nl advisory stubs (read through CSAF and per-year URLs).

← Operations dashboard · run-record contract: docs/pipeline.md