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

2026-11-18T0356Z-intel

One pipeline fire, in full · intel run of 2026-11-18 · 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/2026-11-18/2026-11-18T0356Z-intel.md.

Run telemetry

2026-11-18T0356Z-intel intel prompt v4.25 publish pending
3h 10m duration 10 published 3 updates
Claude Sonnet 5.5 (claude-sonnet-5-5) main agent
S1 Claude Sonnet 5.5 (claude-sonnet-5-5)
Items returned
15
Duration
34m 51s
Tool calls
13 WebFetch16 WebSearch105 bridge
Cited sources
12 of 28 in slice
S2 Claude Sonnet 5.5 (claude-sonnet-5-5)
Items returned
11
Duration
21m 30s
Tool calls
2 WebFetch19 WebSearch100 bridge
Cited sources
6 of 21 in slice
S3 Claude Sonnet 5.5 (claude-sonnet-5-5)
Items returned
9
Duration
24m 50s
Tool calls
7 WebFetch5 WebSearch90 bridge
Cited sources
7 of 16 in slice
S4 Claude Sonnet 5.5 (claude-sonnet-5-5)
Items returned
6
Duration
23m 00s
Tool calls
3 WebFetch36 WebSearch90 bridge
Cited sources
3 of 10 in slice

Verification

#1 NEEDS_FIXES · Sonnet 5.5 · t=15 e=10 a=7 #2 NEEDS_FIXES · Sonnet 5.5 · t=5 e=2 a=3 #3 NEEDS_FIXES · Sonnet 5.5 · t=2 e=2 a=3 #4 NEEDS_FIXES · Sonnet 5.5 · t=3 e=0 a=1 #5 NEEDS_FIXES · Sonnet 5.5 · t=2 e=0 a=0

Deep dive

·

Entries this run published (10) and updated (3)

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.

2 added candidate · 1 last_successful_fetch · 1 notes · 1 consecutive_quiet_periods.

SourceChangeFrom → ToReason
typo3-security-advisoriesadded candidate· → ·primary for TYPO3 core advisories, ten of them on 2026-11-17 with one stated exploited; the typo3.org feed and the advisory pages read with feed and extract
mitel-security-advisoriesadded candidate· → ·MISA-2026-0012 (MiCollab AWV, CVSS 10.0 unauthenticated command injection) read cleanly with extract; no feed found
oracle-cpu, cisa-kev, ncsc-ch-security-hub, enisa-euvd, socket-dev-blog, mandiant-gtig, report-uri-blog, drupal-security, safedep, bleepingcomputer, huntress, unit42last_successful_fetch· → ·fetched and used by an entry or a changelog record published this run
drupal-security, oracle-cpu, ransom-isac, ransomware-live, bleepingcomputer, databreaches-netnotes· → ·working recipes found this run (Drupal PSA through its feed, Oracle CSPU verbose text page, ransom-isac feed.xml, the ransomware.live Swiss victims API, the BleepingComputer listing, databreaches.net paged feed)
(69 slice records)consecutive_quiet_periods· → ·200 but nothing publishable in the horizon: counters incremented by one; the 12 sources used by an entry 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
perforce-portalhttps://portal.perforce.com/s/cve/a91Qi000003IaHpIAK/report-deserialization-codebridge:extract → webfetch → bridge:jina200 js-shell
Salesforce single-page application: extract, WebFetch and the reader all return an empty shell
vendor channel read through Puppet's release notes, the CNA record via ENISA EUVD, the OpenVox advisory and Neodyme
justice-govhttps://www.justice.gov/archives/opa/pr/justice-department-disrupts-russian-intebridge:extract → webfetch403 transport-403
Akamai interstitial on extract, empty body on WebFetch
the FSB statement in the COLD RELIC entry is attributed to GTIG, which paraphrases the DOJ assessment
jina-readerhttps://r.jina.ai/bridge:jina402 reader-pool-dead
HTTP 402 on every pooled key; one fetch was served intermittently
direct transports (extract, url --direct, feed) and WebFetch carried every read

Bridge invocations (this run)

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

6 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 recent and post ×1
  • tools/fetch_source.py pdf ×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 · 32 findings (truth=15, editorial=10, advisory=7) · Claude Sonnet 5.5 · 24m 02s

F-codeSectionItem · URL/quoteVerifier summaryRemediation · outcome
F4
hallucinated-fact
·
Operationally harmful and unsupported: pip list starts the environment's Python, which processes the .pth hook and re-runs the stealer; SafeDep never mentions pip list and warns against starting Pythofixed: Exposure now names lockfiles, requirements files and image build logs, and Detection carries SafeDep's warning to look for the startup file without starting the environment's Python
F4
hallucinated-fact
·
Passkeys, backup codes, rogue devices, persistence across resets and the tenant-wide Customer Takeout export are Case 1 (late 2025) only; Case 2 was one new MFA device and a bulk download of one user'fixed: summary now separates the late-2025 case (push fatigue, enrolled passkeys, backup codes and devices, tenant export) from the 2026 case (phished Entra ID credentials, one new MFA device, one user's drive)
F4
hallucinated-fact
·
The N-able CNA record does not carry the 'before 2026.4.1 HF3' wording for this id (the other sixteen do); it scopes the flaw to the Patch Management Component before 2.14.4.5829. Fix affected/fixed ffixed: affected and fixed now follow the CNA record (Patch Management Component before 2.14.4.5829)
F4
hallucinated-fact
·
CNA vectors carry PR:N and no authentication precondition; the preconditions are AC:H/AT:P (attacker position). post-auth misstates the record; pre-auth (or omit) fits the CNA vector.fixed: auth set to pre-auth for CVE-2026-129286 and CVE-2026-129287, matching the CNA vectors (PR:N, attacker supplies the metadata or files)
F3
claim-not-supported
·
The cited page lists the id among fixes in the 2026.0.1 release; 2025.12.1 and 2023.8.13 for this CVE come from the ENISA EUVD record, not this page. The source date 2026-11-03 for these notes also hafixed: the release notes are cited for 2026.0.1 only, the EUVD record for 2025.12.1 and 2023.8.13, and the release-notes source date is now the page's metadata date 2026-11-12 because the page carries only 'Released…
F3
claim-not-supported
·
Stated for 2026.0.1 only; the entry says 'the upgrade' after offering 2023.8.13, 2025.12.1 or 2026.0.1. Scope the sentence to 2026.0.1 (the OpenVox post says to validate custom auth.conf rules after ufixed: auth.conf sentence scoped to the 2026.0.1 upgrade
F3
claim-not-supported
·
The PSA's own risk vector says only uncommon module configurations are exploitable, which is a configuration condition; the entry cites the 18/25 score but states the opposite for exposure. State the fixed: Exposure and the main text now state the PSA's AC:Complex, E:Proof and TD:Uncommon ratings and its warning that more than one configuration is affected
F3
claim-not-supported
·
The elements trace to GTIG's narrative and rule names, but GTIG does not present them as benign-versus-malicious discriminators. Attribute as 'drawn from GTIG's description' or drop 'GTIG's discriminafixed: Triage reworded to 'the cues in GTIG's account'
F3
claim-not-supported
·
'Many' contradicts the source's 'a handful'. Use the source wording.fixed: 'a handful of repositories', the source's wording
F3
claim-not-supported
·
Second half of the clause is carried by the Trust Center page, not the cited blog. Add the Trust Center citation.fixed: expiry of vendor credentials is now cited to the Trust Center page, the blog keeps the standing-credentials clause
F3
claim-not-supported
·
Derived detection attributed to a page that does not state it. Mark as derived from the mechanism or drop the citation.fixed: the HTML or SVG detection is now cited to SA-032, which states the upload
F3
claim-not-supported
·
The provider-inquiry advice is the entry's own recommendation under a Field Effect citation; split the clause or cite only what the page carries.fixed: the provider-inquiry advice split from the Field Effect citation into its own uncited recommendation
F14
?
·
Oracle names two interim measures; for the unauthenticated rows only the first applies, which the entry should say instead of 'only'.fixed: summary, action and body now say Oracle's interim measure for attacks that need no privileges is blocking the required network protocols, and name the privilege and package-access alternative
F14
?
·
Source counts organisations (and 188 snippets), not sites, and the count is Report URI's, not Optimizely's; the headline juxtaposition reads as Optimizely confirming 178.fixed: headline counts organisations and attributes the count to Report URI
F4
hallucinated-fact
·
No cited source says SSH was the protocol of the attempted connections (credentials are described as SSH private keys and access keys); the mapping infers it.fixed: T1021.004 removed from techniques[]
F8
needs-more-research
·
A cited source reports an independent working-exploit claim (unreleased, application-specific, pre-auth via Basic auth and java-jwt) that changes the exploit-readiness picture for the 'no attacks yet'fixed: Tenable's report of an unreleased working exploit claimed by Immersive's Kevin Breen added to the body, summary and sourcing note, stated as a claim with its limits
F8
needs-more-research
·
Exposure says 'AVX-512 VBMI x86-64 host' without telling a responder which CPUs or how to check, omits the Java 8 performance build that the actions and cves[] name, and Detection has only post-crash fixed: Exposure now names the Java 8 performance build, the required CPU flags and families (Upwind); Detection adds Upwind's request-shape and log signals, the crash-log and restart signals and Talos's Snort coverage…
F8
needs-more-research
·
The entry carries the score only. The vector holds the exploit-status signal (E:Proof) and the exposure limiters (AC:Complex, TD:Uncommon) that a SOC needs for triage; the NCSC.ch 'exploitation unknowfixed: the PSA's risk vector (AC:Complex, E:Proof, TD:Uncommon) is carried in the body
F8
needs-more-research
·
The entry states the MVG file disclosure as reachable 'where such an upload form exists' without the ImageMagick/MVG precondition and omits the post-update image-format configuration the advisory flagfixed: Exposure names the ImageMagick and MVG-coder precondition and the Defender takeaway carries the post-update fileExtensionToMimeType action; the unused pre-announcement source was dropped and the analysis trimmed…
F9
surface-contradiction
·
The body now attributes to Ohey only the Kissmetrics/StorifyMe/minute-ly part and adopts Report URI's date; the disagreement survives only as 'the Ohey figures the entry first relied on' in the updatefixed: the update section now carries a Contradiction on dates naming Ohey's 2026-11-13 and Report URI's 2026-11-16 04:21 UTC first injection, and the repeated Report URI sentences were cut to the delta
F7
drop
·
Out-of-nexus single-victim incident whose exploited flaw is unknown; the transferable lesson (internet-facing status page on a route to operations plus plaintext SSH keys in home directories) is thin.fixed: one-clause relevance ground added (research and high-performance-computing operators, transferable exposure pattern); priority stays notable under the incident floor
F16
?
·
Exploited, patch available, management-plane RCE with a seven-day decision (upgrade, stop autosign, treat the CA key as exposed); may clear the high bar. Constituency use of Puppet is not established declined: held at notable; constituency use of Puppet Enterprise is not established, so the high bar's exposure element is not met; calibration note in the run record
F16
?
·
National CERT attention, a dated decision (update within the window) and anonymous reach suggest high; TD:Uncommon and unknown exploitation argue for notable. Re-grade deliberately and state why.declined: held at notable; the PSA rates attack complexity Complex and target distribution Uncommon and exploitation is unknown; the release's own update record is the point to re-grade; calibration note in the run…
F16
?
·
The update adds a vendor-confirmed unauthenticated 9.8 across current AXIS OS with no fix and a do-now workaround; priority was not re-graded in the record. Re-assess for high.declined: held at notable; the Axis update adds an unauthenticated 9.8 with no fix but no exploitation is reported and the flaw needs HTTP/2 enabled; calibration note in the run record
F18
?
·
The hunt (first clause) is a start-now task; the remediation clause is conditional playbook text duplicating the body. Keep the hunt only.fixed: the second clause of actions[0] removed, the hunt stays
F11
editorial-advisory
·
Fetch transports and reading mechanics in reader-facing sourcing_note; reduce to two sentences of provenance each.fixed: reader-facing sourcing_note wording removed from the Puppet, Drupal, COLD RELIC, agent-framework and N-able notes
F11
editorial-advisory
·
Record-keeping narration in reader text; 'settles' overstates a status line that names no flaw class; the section repeats the body's Optimizely/Report URI sentences nearly verbatim instead of carryingfixed: the update section's record-keeping narration removed, 'settles' dropped, the Report URI repetition cut to the delta
F11
editorial-advisory
·
Changelog summaries describe the entry rather than the finding; state only the delta.fixed: the N-able and Apache update-record summaries now state only the delta
F11
editorial-advisory
·
Neither entry says why it is in a Swiss public-sector brief (Barracuda: supplier and ARTEX technique; RIKEN: see F7). Add the clause.fixed: relevance clause added to Barracuda (supplier incident plus ARTEX-style tradecraft) and RIKEN
F11
editorial-advisory
·
Blanket headline; name the distributions or say 'major distributions ship fixes'.fixed: headline names Debian, Ubuntu and Red Hat
F11
editorial-advisory
·
Contradicts the corrected entry (Optimizely files carried the code from 2026-11-16) and appears on the entity page; refresh the summary.fixed: campaign:optimjack summary now dates the Optimizely files from 2026-11-16 and states Optimizely's confirmation
F11
editorial-advisory
·
N-able N-central is a third high entry touched by the run and is not named; the note should list it.fixed: the calibration note lists the N-able entry

Iteration #2 NEEDS_FIXES · 10 findings (truth=5, editorial=2, advisory=3) · Claude Sonnet 5.5 · 20m 49s

F-codeSectionItem · URL/quoteVerifier summaryRemediation · outcome
F3
claim-not-supported
·
Upwind: 'Untrusted Base64 reaching Base64.getDecoder() or getUrlDecoder()'; Equixly (cited in the same entry): 'only the basic and URL decoders of java.util.Base64 reach it'. The entry's own Upwind-cifixed: Exposure now says the basic or URL decoder of java.util.Base64
F3
claim-not-supported
·
Tenable says only 'The post does not say that the exploit has been released.' That is not a statement that it is unreleased. Reword to 'with no release stated' in the body, summary and sourcing_note.fixed: body, summary and sourcing_note now say a working exploit claimed with no release stated
F3
claim-not-supported
·
BACS says only 'Die Schwachstelle betrifft bestimmte Java-Versionen auf bestimmten, weit verbreiteten Prozessortypen' and names no Java line, no Java 8 performance build and no x86-64; those come fromfixed: Oracle is cited for the affected-version clause, BACS stays on the bundled-runtime and running-build clause
F3
claim-not-supported
·
Equixly says only 'Unexplained JVM crashes, such as new hs_err_pid*.log files ... Do not expect the log to name Base64'. 'bursts of restarts' is Upwind's ('unexplained hs_err_pid*.log files or bursts fixed: Equixly keeps the crash-log statement, the restart bursts moved to the Upwind clause, and the process and administrative-activity hunts are marked as following from Calif's chains
F3
claim-not-supported
·
SafeDep: 'Look for the versions in lockfiles and requirement files. On machines, images and runners, look for either startup file without starting the environment's Python.' It names no image build lofixed: 'image build logs' dropped, SafeDep names lockfiles and requirements files
F8
needs-more-research
·
PSA: 'All versions of Drupal prior to 10.6.x as well as all versions of Drupal 11 prior to 11.3.x are end-of-life ... Sites on those versions should apply the interim mitigation steps'. The Exposure lfixed: Exposure names the end-of-life lines (Drupal before 10.6 and Drupal 11 before 11.3)
F2
generic-url
·
The URL is the advisory listing index, not a specific advisory. The claim is true (latest core advisory in the feed is SA-CORE-2026-016, 2026-10-14) but is a time-sensitive absence that the planned upfixed: the absence sentence and the listing-index source were dropped
F11
editorial-advisory
·
Fetch-transport wording and notes far beyond two sentences of provenance remain in reader-facing sourcing_note fields; reduce to provenance (who is the disclosing party, what is single-sourced, what cfixed: Drupal transport wording removed, the N-able and Apache sourcing notes condensed to provenance
F11
editorial-advisory
·
All four CNA vectors carry PR:H (the entry's own sourcing_note says 'high privileges for CVE-2026-125879, CVE-2026-125880 and CVE-2026-125882'); use one bucket for PR:H records. Not false (admin is aufixed: CVE-2026-125879 and CVE-2026-125882 now use admin-required like the other PR:H records
F11
editorial-advisory
·
The Wiz page ('temporarily disable HTTP/2 support to eliminate the attack vector') still supports the advice and stays in sources[], but no sentence cites it now. Restore a short cited clause or attrifixed: Wiz's advice to disable HTTP/2 temporarily is cited in the Apache Defender takeaway

Iteration #3 NEEDS_FIXES · 7 findings (truth=2, editorial=2, advisory=3) · Claude Sonnet 5.5 · 20m 21s

F-codeSectionItem · URL/quoteVerifier summaryRemediation · outcome
F3
claim-not-supported
·
The BACS-terminated clause carries 'on an x86-64 host' and 'standalone runtimes', which BACS does not state (BACS: 'Die Schwachstelle betrifft bestimmte Java-Versionen auf bestimmten, weit verbreitetefixed: the scope list (standalone, bundled, container, running build) is now cited to the NCSC.ch Hub post, Oracle keeps the version clause and the unsourced x86-64 wording was dropped from it
F3
claim-not-supported
·
SafeDep says 'Look for the versions in lockfiles and requirement files' and 'look for either startup file without starting the environment's Python'; it names no installed-package inventories or CI jofixed: Detection names lockfiles and requirements files (SafeDep) and cites SafeDep for the sign-in-log review, 'unfamiliar addresses' dropped; the action wording now says CI install logs
F5
missing-citation
·
Upwind's current page (the one cited) says the opposite ('It's just a crash. Wrong.') and carries no record of its earlier reading; the only fetched page that states Upwind's early crash reading is Thfixed: the Upwind half of the early-analysis statement was dropped, Equixly alone is named
F8
needs-more-research
·
Field Effect (in sources[] and cited in the body) 'Update 3: November 16, 2026': 'Shortly after the HF3 release, technical details on 24 further N-central vulnerabilities were published. These are repfixed: the 2026-11-18 update section now carries Field Effect's 24 further vulnerabilities reported as addressed by Hotfix 3, stated as not reconciled with the 17 CVE ids
F11
editorial-advisory
·
Unbalanced parenthesis: the '(look for it on disk ...' opened after 'package' is never closed (5 opening, 4 closing parentheses on the line). Close it after the SafeDep citation.fixed: the unclosed parenthesis was removed by restructuring the sentence
F11
editorial-advisory
·
Iteration 2 asked for provenance only; both notes still run well past two sentences (the N-able note lists CNA score sources, the 33-count provenance, the BleepingComputer-versus-Censys date differencfixed: N-able and Apache sourcing_note condensed to provenance and the contradictions
F11
editorial-advisory
·
The PSA's metric is 'AC - Access complexity: Complex or highly specific' (Drupal's own scale, not CVSS 'attack complexity'); say 'access complexity'. Same wording in the run record calibration note (ofixed: 'access complexity Complex', the PSA's own metric name

Iteration #4 NEEDS_FIXES · 4 findings (truth=3, editorial=0, advisory=1) · Claude Sonnet 5.5 · 13m 32s

F-codeSectionItem · URL/quoteVerifier summaryRemediation · outcome
F3
claim-not-supported
·
The sentence inserted in iteration 3 makes Field Effect the subject of the preceding sentence, so 'It advises' now attributes Huntress's four recommendations to Field Effect. Huntress (UPDATE 11/17 @ fixed: the Huntress recommendations in the 2026-11-18 update section now read 'Huntress advises', so the Field Effect sentence is no longer their antecedent
F3
claim-not-supported
·
(low confidence) SafeDep says both versions are 'Malicious, removed from PyPI' and 'The last clean release is 1.26.0'; it never says 1.26.0 is the newest version. The fact is true (PyPI JSON for agentfixed: agent-framework body now says both versions were removed from PyPI and 1.26.0 is the last clean release
F3
claim-not-supported
·
(low confidence) Previdian's page says 'Previdian sensors have observed 19 exploitation attempts, 9 unique attacker IPs'; it does not say honeypot, and Huntback's page never mentions Previdian. 'Honeyfixed: 'Previdian's sensors recorded attempts' in the two places the entry states it
F11
editorial-advisory
·
(low confidence) The central behaviour in both cited sources is a .pth startup hook that runs the stealer every time Python starts (Socket: 'Python startup hooks'; SafeDep: 'A .pth file runs the stealfixed: T1546.018 (Python Startup Hooks, active in the pinned ATT&CK v20.1) added to techniques[]

Iteration #5 NEEDS_FIXES cap-breach · 2 findings (truth=2, editorial=0, advisory=0) · Claude Sonnet 5.5 · 11m 46s

F-codeSectionItem · URL/quoteVerifier summaryRemediation · outcome
F3
claim-not-supported
·
(low confidence) The release notes (cited) name none of the three, and Huntress (cited) says nothing about what N-able's notice names; 'N-able's notice' is a different page (the uptime.n-able.com evenfixed: 'and N-able's notice' dropped from the N-able 2026-11-18 update sentence; the release notes alone are cited
F3
claim-not-supported
·
(low confidence) SafeDep says only: 'Review sign-in and activity logs for the managed identities and federated credentials that environment could use, from the install time onward.' It names no GitHubfixed: agent-framework Detection now uses SafeDep's wording (review sign-in and activity logs for the managed identities and federated credentials the environment could use, from the install time onward)

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

2026-11-18T0356Z-intel · Sonnet 5.5 · window 26 h · 10 entries published

Verification & coverage notes

Window 26 h (gap 24.0 h to 2026-11-17T0355Z-intel). Ten new entries and three changelog records. The will-publish set was too large for a main-agent deep read, so four scoped cti-research follow-ups (FU1 to FU4, Sonnet 5.5, 04:36:43Z to 05:05:26Z) read the primaries and wrote claim sheets under work/2026-11-18T0356Z-intel/deepread.FU*.md; the entries were composed from those sheets and the saved bodies, and each evidence quote was checked as a literal substring of its saved body.

  • Backlog: TYPO3 (published, 2026-11-18/typo3-core-security-releases-2026-11-17-installer-chain) and Oracle CSPU (published, 2026-11-18/oracle-november-2026-cspu-56-unauthenticated-critical-rows) were struck as published; the Open table is empty.
  • KEV: the only in-window addition, CVE-2026-115731 (Perforce Puppet Enterprise), is the entry 2026-11-18/cve-2026-115731-puppet-enterprise-report-deserialization-rce; no covered CVE newly flags ransomware use.
  • Updates: OptimJack (update: Optimizely confirmed an exploited vulnerability; the entry's 2026-11-13 start for the Optimizely files was wrong and is corrected in place), N-able N-central (update: Hotfix 3 fixes 17 more CVE ids, blocks the Legacy API permanently; the "no fix, follow-up hotfix planned" statements are replaced), Apache httpd CVE-2026-119618 (update: Axis OS 6.50.5.10 and higher affected, no fixed release).
  • Single-source: COLD RELIC (Mandiant/GTIG first-hand, no independent coverage), Barracuda and RIKEN (victim's own disclosure; their press relays add nothing independent), Drupal PSA, TYPO3, Mitel and Oracle (disclosing vendor), Puppet (CISA carries exploitation alone, no actor or telemetry; the flaw itself is multi-source).
  • Contradictions surfaced in entries: Java4Shell impact (early source-reading analyses expected a JVM crash, Calif demonstrates code execution, vendors rate CVSS 3.1 8.1 against Calif's 4.0 9.2); TYPO3's "actively exploited" sits on SA-029 (Medium), not on the Critical SA-027 chain, and an automated CISA record says exploitation none; Mitel's range ("10.3.0.18 and earlier") against CERT-FR and BSI ("before 10.3.0.18"); OptimJack's start date for the Optimizely files; Barracuda's ARTEX statement grows in strength across its three pages.
  • First-pass statements the deep reads corrected: a public trigger proof of concept and a CVE record exist for Java4Shell (S3 said neither); the Drupal PSA never says "proof of concept" (the risk vector carries E:Proof); the Axis PDF never says "HTTP/2 flaw on the web interface"; Oracle's "21 families" and "640 CVEs" are Tenable's counts and only the TopLink row is 10.0 (the T3 and IIOP row is 9.8).
  • Lookback item: Barracuda (primary 2026-11-10 to 2026-11-13, first coverage from BleepingComputer's 2026-11-16 sighting within its 49 h lookback; event_date 2026-11-10).
  • Calibration: high on Java4Shell and TYPO3 (OptimJack and the N-able N-central entry stay high); everything else notable. Held at notable after the verifier's iteration-1 challenge: Puppet (exploited per CISA KEV, but constituency use of Puppet Enterprise is not established), Drupal (pre-announcement without CVE, component or fixed version, PSA rates target distribution uncommon and exploit status proof of concept; the update on the release is the point to re-grade) and Apache httpd (Axis finding adds an unauthenticated 9.8 with no fix, but no exploitation is reported and the flaw needs HTTP/2 enabled). No critical: no source shows exploitation of an exposure the constituency demonstrably runs with a time-critical action.
  • Next fire: the Drupal core security release is due 2026-11-18 17:00 to 21:00 UTC, after this run publishes; the heads-up entry needs the SA-CORE advisory (CVE ids, fixed versions, exploitation status) as an update record. Also re-check ZDI advisories for the Oracle APEX and ORDS rows and Huntress's N-central Hotfix 3 exploitation reports.
  • borderline-drop: Apache Roller CVE-2026-129448 (S1.2), CNA-stated exploited but a niche blog server with no constituency exposure established.
  • borderline-drop: Broadcom AutoSys CA20261117-01 (S1.8), internal workload-automation tier, no exploitation, regular patch cycle.
  • borderline-drop: ASUS Control Center hard-coded keys (S1.9), niche, single EUVD record, no exploitation.
  • borderline-drop: Eclipse Mosquitto CVE-2026-110486 (S1.10), websockets listener only, no exploitation, fixed version stated inconsistently (2.1.3 against 2.1.4).
  • borderline-drop: Brotli vendor-impact long tail (S1.14: Siemens ICSA, Oracle rows, Xerox and Lexmark, Cisco revision 1.14); the covered entry already carries the Siemens and Cisco lists and is over its analysis bound; the Oracle rows are in the CSPU entry.
  • borderline-drop: Juniper Junos OS Evolved CVE-2026-98432 (S1.15), single CERT-FR relay, vendor page unreadable, fix pending.
  • borderline-drop: Danfoss AK-SM 800A (S2.3); the covered entry's 2026-11-14 record already carries Midnight Blue, CCB and the BSI position; BSI version 1.1 changes no decision.
  • borderline-drop: BACS weekly review 46 (S2.4), phishing awareness, no new technique.
  • borderline-drop: Operation Endgame Gootloader, CastleRAT and StealC (S2.5), takedown with no defender-side decision change.
  • borderline-drop: FBI and DOJ FaceStealer (S2.6), DPRK crypto-theft social engineering, a class already covered (Binance, Sophos, PolinRider).
  • borderline-drop: Quantum Systems QBase update server (S2.7), press-reported, no Swiss customer, no vector beyond the update server.
  • borderline-drop: Reka (S2.8), Swiss private cooperative, no vector or actor, below the incident floor.
  • borderline-drop: Zurich cloud strategy RRB 1155/2026 (S2.9), cantonal procurement policy with no SOC decision.
  • borderline-drop: Microsoft 365 outage 12 to 13 November (S2.10), availability, not security.
  • borderline-drop: Project Zero "Full X" disclosure policy (S2.11), no near-term defender decision.
  • borderline-drop: Operation ShadowCloud (S3.2), Kaspersky and Sophos corroborate each other, but victims are in Asia and Africa, no source ties it to Europe, and neither post carries the cloud-service detection points the first pass proposed.
  • borderline-drop: ARTEX delta from Amazon MadPot, Cloudflare and Talos (S3.5), financially motivated campaigns against LLM gateways with no government target, mostly roll-up of covered reporting, and the covered entry already exceeds its analysis bound; the Dosu post-mortem that names ARTEX is a lead for the audit.
  • borderline-drop: Turla Cereal delta, Romania's SRI statement (S3.6), names Turla only in an opening clause, says only that unspecified Microsoft system vulnerabilities were exploited, and does not say it is the ESET case.
  • borderline-drop: Fireworks AI two-incident bulletin and Uber (S4.3), no source ties the second incident to the Cursor marketplace compromise; the delta concerns other victims and no Swiss or European public-sector party.
  • borderline-drop: Microsoft device-code phishing against Copilot CLI (S3.8), laboratory proof of concept, not observed, device-code class already covered.
  • borderline-drop: Enea SilentAir and LIMINAL PANDA (S3.9), single vendor, Asian operators.
  • borderline-drop: Infoblox third-party migration breach (S4.4), thin statement, detail rests on anonymous sources relayed by two outlets, no Swiss link.
  • borderline-drop: Kreiskrankenhaus Greiz-Ronneburg (S4.5), no actor, vector or confirmed attack type, below the incident floor.
  • borderline-drop: California v. OpenAI (S4.6), legal milestone, no technical delta.
  • Leads not pursued (below the bar today): N-able Mail Assure and SpamExperts server incident (N-able status page, 2026-11-18 03:05Z, vendor channel, no vector), CNIL's ten fines on French local authorities (2026-11-09, outside the lookback), Maileva (freshest source 2026-11-16, no Swiss client).
  • Included with reduced confidence: none.
  • Coverage gaps: belden-psirt (403), connectwise-trust-center (403), sygnia (403) and lab52 (no route) were not in any slice; perforce-portal (script-rendered page), juniper-psirt (script-rendered page), europol-newsroom (React shell, head metadata only), drupal.org page URLs (client challenge to extract, read through the PSA feed and WebFetch), justice.gov (Akamai interstitial), the Maine and Oregon breach filings behind the Uber count (not read, count not carried), mainichi.jp (paywall).
  • Operational: S3 reported two scratch files written at the filesystem root through a mis-scoped shell variable and moved to its scratchpad; nothing in the repository was touched outside work/. discovered_at stamps were set to the real composition clock. Mid-run stop-hook nudges to commit were not acted on before Phase 6.
  • Verification: five iterations (NEEDS_FIXES at 1 to 5: 32, 10, 7, 4 and 2 findings). Iteration 5 returned two low-confidence F3 findings and no F1 or F4, so the loop ended by the early exit (truth + editorial = 2); both findings were fixed after the pass and not re-verified by a further iteration, hence verification_residual_count: 2. The three F16 priority challenges from iteration 1 were declined and not re-raised. No CLEAN verdict was reached, so there is no double-CLEAN; the entries ship on the early exit.
  • Source health: the sweep left one UNSOLVED row, typo3-security-advisories (candidate added this run), verdict irrelevant: the feed works and is current (newest item 2026-11-17), but its titles carry only two of the health check's relevance terms (RELEVANCE_MIN is 3) because the vocabulary lacks words such as advisory and forgery. Left for the quality audit to widen the vocabulary or accept the verdict; no tool was changed in this fire.

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