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

2026-10-23T0256Z-intel

One pipeline fire, in full · intel run of 2026-10-23 · 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/2026-10-23/2026-10-23T0256Z-intel.md.

Run telemetry

2026-10-23T0256Z-intel intel prompt v4.23 publish ok
3h 00m duration 8 published 4 updates
Claude Sonnet 5.5 (claude-sonnet-5-5) main agent
S1 Claude Sonnet 5.5 (claude-sonnet-5-5)
Items returned
13
Duration
28m 03s
Tool calls
4 WebFetch11 WebSearch150 bridge
Cited sources
10 of 25 in slice
S2 Claude Sonnet 5.5 (claude-sonnet-5-5)
Items returned
7
Duration
23m 12s
Tool calls
3 WebFetch30 WebSearch95 bridge
Cited sources
3 of 20 in slice
S3 Claude Sonnet 5.5 (claude-sonnet-5-5)
Items returned
9
Duration
24m 02s
Tool calls
3 WebFetch7 WebSearch105 bridge
Cited sources
3 of 16 in slice
S4 Claude Sonnet 5.5 (claude-sonnet-5-5)
Items returned
5
Duration
22m 43s
Tool calls
7 WebFetch22 WebSearch95 bridge
Cited sources
2 of 12 in slice

Verification

#1 NEEDS_FIXES · Sonnet 5.5 · t=9 e=7 a=4 #2 NEEDS_FIXES · Sonnet 5.5 · t=6 e=3 a=4 #3 NEEDS_FIXES · Sonnet 5.5 · t=3 e=0 a=2 #4 NEEDS_FIXES · Sonnet 5.5 · t=2 e=1 a=2 #5 NEEDS_FIXES · Sonnet 5.5 · t=1 e=0 a=1

1 entry dropped by verification this run (recorded in the verification & coverage notes).

Deep dive

·

Entries this run published (8) and updated (4)

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.

23 last_successful_fetch · 21 notes · 3 status · 2 fetch_method · 1 consecutive_fetch_failures · 1 added · 1 health_cmd.

SourceChangeFrom → ToReason
enisa-euvdlast_successful_fetch2026-10-22 → 2026-10-23fetched and used
msrc-update-guidelast_successful_fetch2026-10-14 → 2026-10-23fetched and used
fortinet-psirtlast_successful_fetch2026-10-08 → 2026-10-23fetched and used
paloalto-psirtlast_successful_fetch2026-10-22 → 2026-10-23fetched and used
cisa-advisorieslast_successful_fetch2026-10-21 → 2026-10-23fetched and used
bsi-delast_successful_fetch2026-10-17 → 2026-10-23fetched and used
anssi-frlast_successful_fetch2026-10-22 → 2026-10-23fetched and used
mysites-gurulast_successful_fetch2026-09-14 → 2026-10-23fetched and used
splunk-advisorieslast_successful_fetch2026-10-09 → 2026-10-23fetched and used
citrix-netscaler-security-bulletinslast_successful_fetch2026-10-09 → 2026-10-23fetched and used
watchguard-psirtlast_successful_fetch· → 2026-10-23fetched and used
ncsc-ch-security-hublast_successful_fetch2026-10-22 → 2026-10-23fetched and used
heise-seclast_successful_fetch2026-10-20 → 2026-10-23fetched and used
inside-it-chlast_successful_fetch2026-10-21 → 2026-10-23fetched and used
report-uri-bloglast_successful_fetch· → 2026-10-23fetched and used
risky-biz-newslast_successful_fetch2026-10-17 → 2026-10-23fetched and used
securityweeklast_successful_fetch2026-10-13 → 2026-10-23fetched and used
piyologlast_successful_fetch2026-09-12 → 2026-10-23fetched and used
bleepingcomputerlast_successful_fetch2026-10-22 → 2026-10-23fetched and used
huntresslast_successful_fetch2026-10-18 → 2026-10-23fetched and used
kaspersky-securelistlast_successful_fetch2026-09-29 → 2026-10-23fetched and used
esetlast_successful_fetch2026-10-09 → 2026-10-23fetched and used
efk-chlast_successful_fetch2026-10-22 → 2026-10-23fetched and used
hpe-networking-psirtnotes· → recipe note appended2026-10-23 (S1): recipe note from the sub-agent source ledger, text in sources.json
cisa-advisoriesnotes· → recipe note appended2026-10-23 (S1): recipe note from the sub-agent source ledger, text in sources.json
watchtowrnotes· → recipe note appended2026-10-23 (S1): recipe note from the sub-agent source ledger, text in sources.json
esetnotes· → recipe note appended2026-10-23 (S3): recipe note from the sub-agent source ledger, text in sources.json
rapid7-researchnotes· → recipe note appended2026-10-23 (S3): recipe note from the sub-agent source ledger, text in sources.json
sysdignotes· → recipe note appended2026-10-23 (S3): recipe note from the sub-agent source ledger, text in sources.json
stepsecurity-blognotes· → recipe note appended2026-10-23 (S3): recipe note from the sub-agent source ledger, text in sources.json
securityweeknotes· → recipe note appended2026-10-23 (S3): recipe note from the sub-agent source ledger, text in sources.json
push-securitynotes· → recipe note appended2026-10-23 (S3): recipe note from the sub-agent source ledger, text in sources.json
novee-securitynotes· → recipe note appended2026-10-23 (S3): recipe note from the sub-agent source ledger, text in sources.json
air-securitynotes· → recipe note appended2026-10-23 (S3): recipe note from the sub-agent source ledger, text in sources.json
synacktivnotes· → recipe note appended2026-10-23 (S2): recipe note from the sub-agent source ledger, text in sources.json
sekoianotes· → recipe note appended2026-10-23 (S2): recipe note from the sub-agent source ledger, text in sources.json
efk-chnotes· → recipe note appended2026-10-23 (S2): recipe note from the sub-agent source ledger, text in sources.json
cnil-frnotes· → recipe note appended2026-10-23 (S4): recipe note from the sub-agent source ledger, text in sources.json
databreaches-netnotes· → recipe note appended2026-10-23 (S4): recipe note from the sub-agent source ledger, text in sources.json
ransomware-livenotes· → recipe note appended2026-10-23 (S4): recipe note from the sub-agent source ledger, text in sources.json
connectwise-trust-centerconsecutive_fetch_failures· → 1403 on every transport (bridge extract and url, jina, WebFetch); a 403 never demotes
expelstatuscandidate → activecited by published entries from 3 distinct runs (sources.promotion_due)
php-net-newsstatuscandidate → activecited by published entries from 3 distinct runs (sources.promotion_due)
ssd-disclosurestatuscandidate → activecited by published entries from 3 distinct runs (sources.promotion_due)
arista-security-advisoriesadded· → candidateS1 candidate: the vendor PSIRT behind three exploited VeloCloud Orchestrator advisories in 2026
citizen-labnotes· → recipe note appended2026-10-23 (main): source_health needs-content-fix: the RSS feed reads fresh (items to 2026-10-22) but its newest items carry no security v
citizen-labhealth_cmd· → ['extract', 'https://citizenlab.ca/research/']probe the research listing instead of the feed
ec-digital-strategy-newsroomnotes· → recipe note appended2026-10-23 (main): HTTP 403 (CloudFront) on the direct transport, extract, the reader and WebFetch on 2026-10-23, as on 2026-10-22; fetch_m
ec-digital-strategy-newsroomfetch_methodbridge → blocked403 on every transport including WebFetch and the reader on two consecutive days
lab52notes· → recipe note appended2026-10-23 (main): TLS EOF on lab52.io from the container (feed and homepage) on 2026-10-22 and 2026-10-23; `extract https://lab52.io/blog/
lab52fetch_methodrss → jinano direct transport reaches it, the reader does
zimbra-security-advisoriesnotes· → recipe note appended2026-10-23 (main): 503 Scheduled Maintenance page served by wiki.zimbra.com on 2026-10-23 (planned upgrade, after the 502 of 2026-10-22); k

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
connectwise-trust-centerhttps://www.connectwise.com/company/trust/advisoriesbridge:extract → bridge:url → jina → webfetch → websearch403 transport-403
direct HTTP 403 (6068 B); trafilatura no readable body; jina reader proxy relayed an upstream block/challenge; WebFetch: HTTP 403 Forbidden
WebSearch for ConnectWise October 2026 bulletins found only the 2026-09-08 ScreenConnect 26.6.5 bulletin (CVE-2026-84869, in KEV 2026-09-11) and earlier Automate bulletins; the individual bulletin pages under…

Bridge invocations (this run)

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

4 other
  • tools/kev_window_diff.py ×1
  • tools/fetch_source.py cisa-kev ×1
  • tools/fetch_source.py ncsc-csh recent 15 ×1
  • extract, url --direct, jina ×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 · 20 findings (truth=9, editorial=7, advisory=4) · Claude Sonnet 5.5 · 21m 31s

F-codeSectionItem · URL/quoteVerifier summaryRemediation · outcome
F4
hallucinated-fact
·
Live advisory timeline now carries '2026-10-23: Added estimated availability dates (ETAs) for hotfix releases across supported PAN-OS versions' (data-date 2026-10-23T03:45Z, three minutes after the en·
F4
hallucinated-fact
·
Same stale state as the PAN-OS entry: the advisory revised 2026-10-23 (03:45Z) lists hotfix builds with ETAs 10/26-10/30 for PAN-OS 12.2, 12.1, 11.2, 11.1 and 10.2; only Panorama and WF-500 are undate·
F3
claim-not-supported
·
Page now reads: 'as of 7 PM ET on October 22, we've identified at least one PoC that successfully bypassed ASLR on x86-64 systems for a single version of PHP (8.4.26) with default heap configurations.·
F4
hallucinated-fact
·
https://watchtowr.com/intelligence/php-cve-2026-109618-faq/ now reads 'Exploitation has been observed in the wild since October 21, 2026, through watchTowr’s Attacker Eye global honeypot network.' (al·
F3
claim-not-supported
·
Both cited pages date the Apple block 2026-10-22: tracker 'blocked in Firefox and Chrome since the 17th of October, on Microsoft platforms since the 20th and on Apple products since the 22nd'; Bleepin·
F3
claim-not-supported
·
The tracker (as saved in the run's bodies/u3-emudhra-tracker.txt and live) lists Node.js 26.12.1/24.22.1/22.23.4, Bun 1.4.4 and FreeBSD ca_root_nss 1.131.1 / caroot patches as fixed; the 'Pending' val·
F3
claim-not-supported
·
(low confidence) The cited blog is dated 2026-10-21 and says 'On October 22, 2026, Fortinet will release a coordinated set of security updates' (future tense); heise (2026-10-22) says the builds 'soll·
F4
hallucinated-fact
·
(low confidence) The hardening post's FortiMail rows (8.0.0-8.0.1 -> 8.0.2, 7.6.0-7.6.6 -> 7.6.7, 7.4.0-7.4.8 -> 7.4.9, 7.2.0-7.2.9 -> move to 7.4) are identical to the fix table of FG-IR-26-175 (CVE-·
F3
claim-not-supported
·
(low confidence, paywalled) The readable lede and bullets say only 'Das betroffene Softwareunternehmen meldete im Januar, es habe die Informationen über Ex-Bundesräte und Agenten gelöscht. Doch dies w·
F6
strengthen-primary-source
·
The cited page is a structured re-write of another outlet's report (its footer reads 'Quelle: ticinonews.ch'), with table-and-bullet filler; the originating article is readable: https://www.ticinonews·
F7
drop
·
(low confidence) No constituency nexus is shown (no Swiss or public-sector deployment evidence), no exploitation or public PoC is reported ('No source fetched reports exploitation'), and the only urge·
F7
drop
·
(low confidence) CVE-2026-109618/109619 are a different finding from CVE-2026-76466 and the Cisco hardening releases; the entry's title, analysis and Exposure/Detection lines still describe only the c·
F9
surface-contradiction
·
(low confidence) Inside IT (2026-10-22), citing Tamedia, says 'Die Bundesanwaltschaft habe bestätigt, dass Daten von der Ransomware-Gruppe Akira abgezogen wurden'; the Federal Prosecutor's Office is a·
F9
surface-contradiction
·
(low confidence) The bulletin prose lists those five branches; the CSAF known_affected ranges are wider: '>=10.8.0.0|<=10.8.1.1', '>=10.6.0.0|<=10.7.2.6', '>=10.3.0.0|<=10.4.1.13', '>=8.12.0.0|<=8.13.·
F10
missed-angle
·
The run record lists both as 'outside this window and with no store entry, for the audit's re-sweep'; I confirmed both pages (published 2026-10-13, Critical, unauthenticated) and that neither CVE is i·
F18
action-item-discipline
·
Restates the entry's Detection paragraph and the last sentence of immediate_action ('check web roots, logs, crashes and unexpected processes') as a task (F18(b)/(d)); the only new element, the 2026-10·
F11
editorial-advisory
·
Process language ('was read', 'in any source read', 'The entry's analysis now also carries') in sourcing_note and in a reader-facing section; the last PHP sentence is record-keeping narration (4c(i)) ·
F11
editorial-advisory
·
A literal service name is a file-name-class indicator; the Detection line already works with 'a service with an updater-style name'. Drop the literal name from prose and evidence.·
F11
editorial-advisory
·
Kaspersky attributes BlackX's exploitation to Baton Drop (CVE-2022-21894) and mentions CVE-2023-24932 only as the vulnerability ESET's post said an unspecified UEFI component 'might exploit'; Microsof·
F11
editorial-advisory
·
'is hit by' reads as exploited while the summary says Citrix does not say NetScaler has been attacked (use 'is affected by'); Huntress's figures are 12 and 4 minutes, the unit is missing; the 2026-10-·

Iteration #2 NEEDS_FIXES · 13 findings (truth=6, editorial=3, advisory=4) · Claude Sonnet 5.5 · 21m 48s

F-codeSectionItem · URL/quoteVerifier summaryRemediation · outcome
F3
claim-not-supported
·
(low confidence) 'the supplier, PK Softech AG of Reinach (BL), says unknown persons used malware...' is cited to the PK Softech notice, which does not name Reinach or Basel-Landschaft (it says only 'E·
F14
quantifier-without-source
·
'first' is supported by no cited source (the clause terminates on Publica's page, which says only 'Abklaerungen dauern noch an') and the entry's own '## Update 2026-10-21' already reports 'the involve·
F3
claim-not-supported
·
(low confidence) The advisory lists three conditions for the Threat IDs ('Route incoming traffic for the MGT port through a DP port', 'Replace the Certificate for Inbound Traffic Management', 'Enable ·
F3
claim-not-supported
·
(low confidence) Cited to FG-IR-26-185 and FG-IR-26-180. FG-IR-26-185 lists no workaround at all, and FG-IR-26-180 says only 'Restricting access on tcp port 27918 prevents the issue.' Neither page say·
F4
hallucinated-fact
·
(low confidence) Frontmatter marks CVE-2023-24932 exploited, but the body (correctly) says 'Kaspersky ties the exploitation to Baton Drop alone; CVE-2023-24932 ... appears in its post only as the one ·
F4
hallucinated-fact
·
(low confidence) The notes say the YOOtheme entry was 'written, then cut at verification iteration 1' and the delta file says the file was removed, yet the counter is 0. Also 'completed: 2026-10-23T03·
F9
surface-contradiction
·
(low confidence) The main analysis states this in the present tense with no date, while the newest section reports (second-hand) that 'the Federal Prosecutor's Office confirmed that data was extracted·
F9
surface-contradiction
·
(low confidence) The advisory page says CWE-89 SQL injection leading to operator privileges ('access to functionality that is intended for internal use only'), while the CSAF cited by the entry carrie·
F8
needs-more-research
·
(low confidence) The cited tracker lists the fixed Java builds ('Java | 27.0.1.0.2, 25.0.5.0.2, 21.0.13.0.2, 17.0.21.0.2, 11.0.33.0.2, 8u511') while Oracle's alert lists 27.0.1, 25.0.5, 21.0.13, 17.0.·
F11
editorial-advisory
·
Store-internal wording in reader-facing prose (check 12). Name the earlier advisory or Defused report and cite it instead of pointing at 'the store' or 'the earlier entry'.·
F11
editorial-advisory
·
The Update section repeats the main-analysis Java paragraph almost fact for fact (Oracle CVSS and exploitability, OpenJDK alternatives, distribution packages, NCSC.ch status). Keep the delta in the se·
F11
editorial-advisory
·
'available to cite' is verification-process wording in reader-facing text (summary, sourcing_note, section); say 'neither has made a statement'. The headline states as fact what the Tamedia lede ('off·
F11
editorial-advisory
·
'hits' reads as exploitation while Palo Alto 'is not aware of any malicious exploitation'; iteration 1 changed 'is hit by' to 'is affected by' on three other headlines. Use 'affects'.·

Iteration #3 NEEDS_FIXES · 5 findings (truth=3, editorial=0, advisory=2) · Claude Sonnet 5.5 · 17m 12s

F-codeSectionItem · URL/quoteVerifier summaryRemediation · outcome
F3
claim-not-supported
·
(low confidence) 'Netzwoche, relaying the same paper, adds that the Federal Council and the security authorities now assume data on practically all current and former federal and federally affiliated ·
F3
claim-not-supported
·
(low confidence) 'NCSC.ch's advisory of 2026-10-22 lists the exploitation status as unknown and the Oracle and OpenJDK guidance as the temporary mitigation'. NCSC post 13091 says 'Temporary: Block the·
F14
quantifier-without-source
·
(low confidence) Huntress: 'Look for rules with one to three-character or punctuation-only names' and its table lists the rule names ., sss, ss, iT, ',.' and '........'; only some are one character. T·
F11
editorial-advisory
·
Advisory, may be left. The body describes behaviors that ESET maps and that have no id: files and screenshots exfiltrated (T1113), a 10-minute startup delay (T1678, 'Some samples also include a 10-min·
F11
editorial-advisory
·
Advisory, may be left. The only cited source is the Citrix post, which calls the flaws 'two PHP memory safety vulnerabilities' and never uses the name PHP2Shell; the name is in the referenced PHP entr·

Iteration #4 NEEDS_FIXES · 5 findings (truth=2, editorial=1, advisory=2) · Claude Sonnet 5.5 · 19m 57s

F-codeSectionItem · URL/quoteVerifier summaryRemediation · outcome
F3
claim-not-supported
·
(low confidence) The entry says 'no exploitation or public code known' (headline), 'HPE publishes no detection guidance and states no exploitation' and 'HPE reports no exploitation'. The bulletin's 'E·
F4
hallucinated-fact
·
(low confidence) The sentence is uncited and the entry's own newest section contradicts it. Inside IT 2026-10-22 (https://www.inside-it.ch/publica-wollte-pensionskassendaten-loeschen-lassen-20261022):·
F5
missing-citation
·
(low confidence) A general claim with no inline citation, and overbroad against the page cited right after it: OpenJDK's advisory says 'If you use an OpenJDK Java runtime provided by your operating sy·
F11
editorial-advisory
·
Advisory, may be left. The round-3 additions (T1573, T1113, T1678) are in and active. ESET's own ATT&CK table (v19) still maps T1090.001 (SOCKS5 server) and T1090.003 (Yggdrasil C2) where the entry ca·
F11
editorial-advisory
·
Advisory (low confidence), may be left. The new wording matches Huntress's own phrase 'one to three-character or punctuation-only names', but drops 'or punctuation-only': Huntress's rule table include·

Iteration #5 NEEDS_FIXES cap-breach · 2 findings (truth=1, editorial=0, advisory=1) · Claude Sonnet 5.5 · 18m 00s

F-codeSectionItem · URL/quoteVerifier summaryRemediation · outcome
F3
claim-not-supported
·
(low confidence) The second half is right (Netzwoche relays the Tages-Anzeiger and quotes no Confederation, Publica or PK Softech confirmation), but the clause 'from unnamed sources' is not on the cit·
F11
editorial-advisory
·
Advisory, may be left (declined last round). The iteration-4 fix is correct (T1090.001 and T1090.003 are ESET's own sub-technique ids and active in the pinned v20 data). ESET's ATT&CK table also maps ·

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

2026-10-23T0256Z-intel · Sonnet 5.5 · window 26 h · 8 entries published

Verification & coverage notes

Window: gap 24.0 h from 2026-10-22T0256Z-intel, window 26 h, developing-story window 72 h. The container clock agreed with the network date (skew 0 s) and the first fetch of origin/main was current. A Friday-morning fire in the second day of the PHP2Shell story: the new signal is downstream (vendor impact statements, the Swiss and UK public-sector picture, diverging exploitability assessments), a vendor-confirmed exploited SD-WAN controller flaw, two out-of-cycle vendor releases (HPE Aruba, Fortinet), and three identity, boot-chain and C2 tradecraft reports.

KEV sweep (work/2026-10-23T0256Z-intel/kev-window.txt, catalogue 2026.10.21): no KEV addition inside the window, nothing NOT COVERED, no RANSOMWARE row; S1 read the live catalogue three times and confirmed it. Neither PHP CVE and not CVE-2026-114625 is KEV-listed.

Backlog (state/coverage_backlog.md): no open row, nothing carried or added; the Beyond Gravity row stays struck because S2 found nothing newer than 2026-10-15.

Published (8 new): cve-2026-114625-arista-velocloud-orchestrator-sqli-exploited (high: vendor-confirmed exploitation, any tenant login, no KEV listing; Arista is first-party and the only source, hence single-source; the two earlier VeloCloud entries are declared in references[]), hpesbnw05162-hpe-aruba-aos-8-aos-10-72-cves-19-unauth (notable: 19 unauthenticated network rows at 9.0 or above in one bulletin, HPE knows of no exploitation or public code; clears PD-11(b) through the number of unauthenticated rows on a management plane and campus wireless estates, a one-clause judgement), netscaler-php2shell-management-gui-fixes-2026-10-27 (notable: own entry because the product, the surfaces and the fix date are NetScaler's; declares the PHP entry and the PAN-OS entry in references[]; lookback item, source dated 2026-10-21 20:23 UTC), fortinet-hardening-release-fortios-fortimanager-fortimail (high: out-of-cycle critical and high fixes the vendor urges at once, advisories withheld to 2026-11-10, exploitation unknown; techniques[] carries T1190 on the vendor's statement that internet-facing devices come first and the interim measure restricts administrative interfaces, no CVE ids exist yet), fortinet-fortisandbox-fortisiem-pre-auth-critical-fixes (notable: first coverage of the Fortinet PSIRT bundle of 2026-10-13, found through the verifier's coverage pass, two unauthenticated critical rows, FortiSandbox GUI authentication bypass at 9.6 and FortiSIEM 7.5.0 path traversal to code execution at 9.4, neither listed as exploited; the FortiSandbox ranges include the builds that fixed the three flaws reported exploited in June, which is the non-obvious defender hook; references the June FortiSandbox entry and the hardening entry), m365-token-spray-replay-rogue-entra-devices-mfa-wipe (notable: Huntress only, behaviour-rich identity tradecraft against the Microsoft 365 estate the constituency runs), blackx-uefi-bootkit-baton-drop-revocation-default-off (notable: a state-actor bootkit whose enabling condition is a default-off Microsoft control; actions[] empty because the revocation is a fleet programme, not a do-now task; references the FishMonger and UEFI-shim entries because CVE-2023-24932 appears in the store) and mythicratatoskr-yggdrasil-mesh-c2-uac-0099-sandworm-wiper (notable: Ukrainian victim of 2025, kept for the mesh-C2 detection hook and the access-handoff-to-wiper pattern; ground (d)). Updated (4, all type: update): the PK Softech entry (the supplier's January deletion pledge apparently not kept, a second-hand report that the Federal Prosecutor's Office confirmed Akira, the Ticino fund IPCT), the PHP2Shell entry (BACS: no known Swiss attack but federal PHP applications affected, the UK Lords figures, AWS's 2026-10-17 first attempt as relayed, Sansec's and Red Hat's cautious exploitability assessments against VulnCheck's exploit, vendor impact statements; priority left at critical because attack traffic is observed, authorities order the update and VulnCheck holds a working exploit, with no compromise confirmed), the eMudhra entry (Java: Oracle alert and OpenJDK advisory for CVE-2026-91403, NCSC.ch 13091) and the PAN-OS entry (the advisory's 2026-10-23 hotfix estimates for PAN-OS 12.2 to 10.2, WildFire WF-500 and WF-500-B and Panorama still undated). PHP2Shell on Cisco Expressway stays a one-clause statement on the PHP entry: it is a different finding from the CVE-2026-76466 entry, so no changelog record was written there.

Borderline drops: borderline-drop: YOOtheme Pro 5 image handler CVE-2026-102785 — unauthenticated CVSS 4.0 10.0 but no exploitation reported and no constituency deployment shown; a vendor-ordered immediate update and a package-diff write-up do not clear PD-11 (written, then cut at verification iteration 1); borderline-drop: Splunk October bundle SVD-2026-1007 to 1013 — scheduled bundle, grouped hardening rows with no stated prerequisites, no exploitation, regular patch cycle; borderline-drop: ISC BIND CVE-2026-94300 and two DoS flaws — CVSS 8.1 with high complexity, needs a captured signed GSS-TSIG message, no exploit known; borderline-drop: Roundcube 1.6.20 and 1.7.5 — about twenty fixes, authenticated paths, no exploitation statement; borderline-drop: Chrome 156 and the Edge Chromium batch — no exploitation reported, routine browser cycle; borderline-drop: MikroTik CVE-2026-89035 — unreviewed CVE record plus a BSI marker, no vendor advisory, so a no-fix claim cannot be cited to the vendor; borderline-drop: Siemens October Patch Day via CISA ICSA-26-295 — no exploitation, no constituency deployment shown; borderline-drop: EFK audit of the federal Microsoft 365 tenant — oversight audit, no near-term defender decision; borderline-drop: Swiss Government Cloud tender — procurement, non-security; borderline-drop: Beyond Gravity — no development since 2026-10-15, the struck row's re-open condition is unmet; borderline-drop: Ebury 1.8.3 on 333 Tor exit relays — no constituency nexus (relay operators and hosters); borderline-drop: CVE-2026-99999 PwnEscape (Linux KVM guest-to-host) — needs an attacker-controlled guest, no exploitation, no workaround, regular kernel cycle; borderline-drop: DeadTag expired-tag domains — vendor-driven research, no government nexus, the exploit-kit leg was dropped by earlier fires; borderline-drop: Rapid7 Q3 2026 quarterly report — statistics without a defender takeaway; borderline-drop: Hunt Intelligence Graylog and X-Road activity in Ukraine — Ukraine-only victims, Log4Shell public-PoC chain; borderline-drop: P3 and Navigate360 anonymous-tip breach — no Swiss customer shown, out of nexus; borderline-drop: Railway environment-variable exposure — no vector or actor yet and no Swiss nexus, preliminary report due 2026-10-23 midday ET, a later fire re-gates it; borderline-drop: NEC sub-contractor credential abuse (Japan) — out of nexus, lookback item; borderline-drop: Capcom, Onamae Desktop Cloud, the AfD mail leak and the US, Australian and Japanese incidents S4 listed — out of nexus, no transferable technique; borderline-drop: Ircantec forum-sale claim and Medienwerkstatt (Akira claim, mis-tagged DE) — claims without victim confirmation, PD-6; worth a re-check next fire. out-of-window: Sophos IR vishing plus SSPR abuse with a Primary Refresh Token session — primary dated 2026-10-19, window_hours=26, not a rotational-lookback item; the audit may want it, as it may Talos "Parasitizing trust" (2026-10-21, trend piece, dropped for recency and novelty).

Single-source items and carve-outs: the Arista, HPE, NetScaler and two Fortinet entries are first-party vendor advisories for their own products (NCSC.ch and heise restate Fortinet and add no independent assessment); Huntress, Kaspersky (the campaign and attribution, with Microsoft's KB independently confirming the default-off revocation) and ESET are single research sources; the Arista exploitation statement has no independent corroboration yet. Contradictions and unconfirmed claims reported, not resolved: Inside IT relays Tamedia that the Federal Prosecutor's Office confirmed Akira exfiltrated the data, Publica's own page of 2026-10-20 still says it is clarifying, and no prosecutor text was found; Sansec and Red Hat judge PHP2Shell exploitation harder than VulnCheck's weaponized exploit shows; Huntress gives 303 successes in 29 organizations and later 232 in 28 for one address, so no count from that passage is used.

Coverage gaps: connectwise-trust-center (403 on every transport including WebFetch and the reader, third consecutive run; ConnectWise coverage rests on search and the CCCS feed). Outside this window and with no store entry, for the audit's re-sweep: SAP Security Patch Day 2026-10-13 and the two non-exploited publicly disclosed October Microsoft zero-days. The Fortinet PSIRT bundle of 2026-10-13 (FortiSandbox, FortiSIEM) was first covered by this run after the verifier flagged it. Tages-Anzeiger has no full-text transport (lede and bullets only), so the Tamedia reporting rests on relays. The jina reader pool had three of four keys at HTTP 402 (operational note only). sources.promotion_due was acted on: expel, php-net-news and ssd-disclosure moved to active.

Candidate sources: one added, arista-security-advisories (the PSIRT behind three exploited VeloCloud Orchestrator advisories this year; extract plus the CSAF route). Suggested and not added under the one-per-run guide: isc-bind-security-matrix, ukauthority, 20min-ch, tio-ch, pwn.ai, CBC News Investigates; novee-security and air-security are noted for demotion at the next audit.

Entity registry: four entities registered (campaign:m365-token-spray-replay-huntress-2026, malware:blackx-uefi-bootkit, actor:apt41, malware:mythicratatoskr; attributed-to, affects relations sourced to the entries that state them) and eight product entities synced. BlackLotus is named only as a comparison and stays unkeyed.

Self-identification: this fire's main agent is Sonnet 5.5 (claude-sonnet-5-5) per the harness model line; the four research sub-agents reported Sonnet 5.5 (claude-sonnet-5-5) in their own return headers.

Verification loop: five cti-verification passes, all NEEDS_FIXES, with truth plus editorial counts of 16, 9, 3, 3 and 1 and every claim of every pass walked against the ledger (303, 282, 222, 284 and 165 claims in scope, all answered). Iteration 1 found the stale PAN-OS state (the advisory gained hotfix estimates three minutes after the entry was written), a VulnCheck sentence the page had since reversed, two stale eMudhra statements, the FortiMail overlap with the exploited CVE-2026-104286 entry, a Frontaliere Ticino aggregator citation, the YOOtheme entry (dropped: no constituency nexus, no exploitation), a PHP2Shell update on the Cisco Expressway entry that did not belong to that finding (reverted, one clause stays on the PHP entry), and the Fortinet PSIRT bundle of 2026-10-13 (new entry). Iterations 2 to 4 removed overstatements the fixes themselves introduced or exposed (a location and a decryption condition no source states, an uncited workaround, an exploited status on CVE-2023-24932, undated present-tense statements, store-internal wording, an eMudhra Update section that repeated the main analysis, the HPE exploitation attribution). Iteration 5 left one low-confidence F3 (from unnamed sources, fixed after the pass and not re-verified) and one advisory (extra ATT&CK ids on the Mythic entry, one added). The loop ended by the early exit of the decision rules (truth plus editorial at most 2, no F1 or F4), so verification_residual_count is 1 and no CLEAN confirmation applies; entries_dropped_by_verification is 1 (YOOtheme). One advisory finding was declined with a rebuttal: the NetScaler entry keeps the name PHP2Shell, which Sansec, VulnCheck, watson.ch and The Register use for these flaws, while its own cited source (Citrix) says two PHP memory safety vulnerabilities.

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