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

2026-11-17T0355Z-intel

One pipeline fire, in full · intel run of 2026-11-17 · 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/2026-11-17/2026-11-17T0355Z-intel.md.

Run telemetry

2026-11-17T0355Z-intel intel prompt v4.25 publish ok
2h 09m duration 6 published 2 updates
Claude Sonnet 5.5 (claude-sonnet-5-5) main agent
S1 Claude Sonnet 5.5 (claude-sonnet-5-5)
Items returned
10
Duration
29m 18s
Tool calls
11 WebFetch15 WebSearch120 bridge
Cited sources
4 of 21 in slice
S2 Claude Sonnet 5.5 (claude-sonnet-5-5)
Items returned
7
Duration
22m 53s
Tool calls
5 WebFetch29 WebSearch75 bridge
Cited sources
1 of 27 in slice
S3 Claude Sonnet 5.5 (claude-sonnet-5-5)
Items returned
9
Duration
19m 01s
Tool calls
3 WebFetch4 WebSearch78 bridge
Cited sources
1 of 15 in slice
S4 Claude Sonnet 5.5 (claude-sonnet-5-5)
Items returned
7
Duration
29m 50s
Tool calls
7 WebFetch34 WebSearch125 bridge
Cited sources
1 of 10 in slice

Verification

#1 NEEDS_FIXES · Sonnet 5.5 · t=11 e=2 a=7 #2 NEEDS_FIXES · Sonnet 5.5 · t=3 e=2 a=3 #3 NEEDS_FIXES · Sonnet 5.5 · t=2 e=1 a=2 #4 NEEDS_FIXES · Sonnet 5.5 · t=0 e=1 a=1

Deep dive

·

Entries this run published (6) and updated (2)

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 rss_url · 1 notes · 1 fetch_method · 1 consecutive_fetch_failures · 1 consecutive_quiet_periods.

SourceChangeFrom → ToReason
ohey-caadded candidate· → ·dated first-party web-threat research with its own captures; the OptimJack timeline and the iPhone-router reading come from it, and it has no feed (listing through url --direct, articles through extract)
aws-security-blogadded candidate· → ·first-party Amazon Threat Intelligence research with a verified feed; its 2026-11-13 ARTEX decoy report was dropped as marginal, so the record starts unused
infostealers-hudson-rock, symantec-security-com, heise-sec, fortinet-psirt, cccs-alerts, vulncheck, enisa-euvd, unit42, huntress, mandiant-gtiglast_successful_fetch· → ·fetched and used by an entry or a changelog record published this run
connectwise-trust-centerrss_url· → ·the bulletin page answers 403 on every direct transport, while the ConnectWise-Advisories GitHub Atom feed reads and carries the same disclosures
symantec-security-com, ico-uk, connectwise-trust-center, enisa-euvd, ncsc-ch-security-hub, ransomware-live, fuitesinfos, ssd-disclosure, msrc-update-guide, cisa-directivesnotes· → ·working recipes found by the sub-agents (Symantec feed, ICO sitemap subcommand, ConnectWise Atom, EUVD search API, security-hub edit history, ransomware.live RSS, fuitesinfos listing, ssd wp-json, MSRC revised sweep, undated CISA directive listing)
anssi-cyber-gouv-actualitesfetch_method· → ·health flagged needs-content-fix: the RSS feed and the listing answer an Incapsula challenge on every direct transport and extract falls through to the dead reader pool; url --direct on the sitemap reads 421 dated URLs (113 under /actualites/, newest 2026-11-12), so health_cmd now runs the sitemap and the record is pinned to the bridge
belden-psirt, sygnia, lab52consecutive_fetch_failures· → ·transport failures on every route (no demotion: 403 or anti-bot blocks never demote)
(63 slice records)consecutive_quiet_periods· → ·200 but nothing publishable in the horizon: counters incremented by one

Coverage gaps (this run)

Sources this run's brief needed that returned no usable content via any documented recipe. Bridge-recovered or quiet-day sources do NOT appear here. (Distinct from the independent source-accessibility probe at the foot of this section, which probes all active sources regardless of what any run needed.)

Source (uncovered)URL triedMethod chainStatus / classWhat the agent did instead
belden-psirthttps://www.belden.com/securitybridge:url --direct → bridge:extract → websearch403 transport-403
403 on every direct transport; the reader pool answers 402 on every key and 401 on the anonymous tier
none; WebSearch found no November 2026 Belden advisory either
sygniahttps://www.sygnia.co/blog/bridge:extract → bridge:url --direct → webfetch → websearch403 transport-403
403 on the blog, feed and sitemap through every transport, seventh consecutive run
none; the record stays pinned blocked
lab52https://lab52.io/blog/feed/bridge:feed → bridge:extract → bridge:url --direct → webfetchn/a transport-timeout
feed and extract fell through to the dead reader pool, url --direct was reset, WebFetch returned 503; fifth consecutive failed run
none; no in-window Lab52 item could be identified

Bridge invocations (this run)

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

5 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 ×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 · 20 findings (truth=11, editorial=2, advisory=7) · Claude Sonnet 5.5 · 19m 24s

F-codeSectionItem · URL/quoteVerifier summaryRemediation · outcome
F3
claim-not-supported
·
Capability and date are spread across all four providers: the clipper and skimmer belong to the Optimizely track.js payload from 2026-11-16, while Kissmetrics, StorifyMe and minute-ly deliver the routfixed: title, headline, summary, body, actions[0] and the registry summary now attribute the router loader to Kissmetrics, StorifyMe and Optimizely from 2026-11-13 (minute-ly replaced outright) and the clipper and…
F3
claim-not-supported
·
(low confidence) "no provider statement was found" is a negative search result attached to a sentence citing Hudson Rock, which does not say it.fixed: removed from the cited sentence; kept in the sourcing note as a statement about the time of writing
F3
claim-not-supported
·
(low confidence) The Ohey citation date 2026-11-17 is the last-read date of a living page whose own date is 2026-11-05.fixed: cited at the page date 2026-11-05 and the sourcing note says the page text runs to 2026-11-17
F3
claim-not-supported
·
The HTTP/2 fallback is stated only for the analysed PerfectPioneer variant; KUKURAT ships separate HTTP/2 and HTTP/3 builds.fixed: sentence scoped to the analysed variant and the separate KUKURAT builds
F3
claim-not-supported
·
(low confidence) The isDangerousCommand blocklist is in two of six PerfectPioneer variants, not the family.fixed: entry and registry summary now say two of the six variants (Pioneer and rat_http3_cc)
F3
claim-not-supported
·
(low confidence) The stealth-for-velocity assessment closes GTIG's section on open-source pentest frameworks, not exploit_agent, and "high-rate scanning" is not in the text.fixed: re-attributed to the pentest frameworks in its own sentence and the Detection line reworded to the traces GTIG names
F14
?
·
(low confidence) "adds three findings" attributes a count GTIG does not state.fixed: reworded so the count is the entry's selection, not GTIG's
F13
?
·
The HVCI-disabling WinRing0 driver link is the entry's inference; Huntress lists the driver in the package separately.fixed: the package also drops the driver, without binding it to HVCI (pre-existing sentence)
F3
claim-not-supported
·
Update record summary splices the 688,000 count onto the 2026-11-12 compromise; the count spans 2026-10-20 to 2026-11-16.fixed: record amended to the correct window and the title field added
F3
claim-not-supported
·
(low confidence) actions[2] dates the nine seconds from the join; Huntress dates it from the file run.fixed: wording now says after its client ran a guest-pushed file
F14
?
·
(low confidence) "two and a half weeks before any advisory existed" matches neither the 3 September advisory nor the 8 September bulletin.fixed: now since 20 August, more than two weeks before ConnectWise's 8 September disclosure (pre-existing sentence)
F9
surface-contradiction
·
(low confidence) An aggregator reports password sign-in restored the same afternoon behind an e-mail code; the entry stated the shutdown as current.fixed: headline dated to 16 November, aggregator report added with its citation and a no-primary caveat, takeaway worded as reported
F18
?
·
actions[2] restates the update section's detection paragraph at about 100 words.fixed: shortened to one do-now task and the artefact list left to the body
F11
editorial-advisory
·
The virtual-patch sentence in the update section is not a delta (the CVE record already named it).fixed: sentence removed; the hardening sentence carries the citation
F11
editorial-advisory
·
Em dashes in the title and body of an entry modified this run.fixed: replaced with commas, colons and parentheses
F11
editorial-advisory
·
(low confidence) The sourcing note ran to three sentences with an analytic link.fixed: folded into two sentences, keeping the disclosed alias link
F11
editorial-advisory
·
The headline stated the DPRK attribution flatly although GTIG writes suspected.fixed: headline now says suspected
F11
editorial-advisory
·
(low confidence) The actively-exploited tag and exploited status are a step stronger than the CCCS and VulnCheck statements.left: advisory; body, summary and sourcing note attribute the claim, and the status rests on a national-authority statement
F11
editorial-advisory
·
(low confidence) The entry did not state why a single US retail case is carried.fixed: sentence added naming the transferable chain and the AI-assisted tempo
F11
editorial-advisory
·
(low confidence) The registry summary of the OptimJack campaign carried the same conflation and a stronger delivery claim.fixed: registry summary rewritten

Iteration #2 NEEDS_FIXES · 8 findings (truth=3, editorial=2, advisory=3) · Claude Sonnet 5.5 · 14m 08s

F-codeSectionItem · URL/quoteVerifier summaryRemediation · outcome
F3
claim-not-supported
·
(low confidence) The summary says Hudson Rock matched archived responses on 21 named sites; the archive check covers 15 captures and the 21 properties combine direct recovery with it.fixed: summary now says Hudson Rock names 21 web properties
F3
claim-not-supported
·
(low confidence) The Exposure clause "not the provider's console" is attached to a Hudson Rock citation that never mentions a console.fixed: clause replaced by Hudson Rock's own advice to verify what visitors actually receive
F3
claim-not-supported
·
(low confidence) The Ohey date 2026-11-05 is an extractor guess from a body sentence and sits on events dated 13 to 17 November.fixed: citations carry 2026-11-17, the latest date the page text carries, and the sourcing note describes the page as a living reference
F9
surface-contradiction
·
(low confidence) Unit 42's appendix says two PerfectPioneer variants rely on HTTP/2 alone, so the HTTP/3-primary generalisation and the UDP 443 detection miss them and the KUKURAT HTTP/2 build.fixed: body names the HTTP/2-only variants, the summary and registry say most variants, and Detection pairs UDP 443 with proxy and TLS egress telemetry
F9
surface-contradiction
·
(low confidence) The aggregator attributes the credential-stuffing label to the BSI while the entry says it is Handelsblatt's, and the title kept the shutdown in the present tense.fixed: the conflict is stated with a citation and the title is past tense and dated
F11
editorial-advisory
·
Three findings reads as a GTIG count although it is the entry's selection.fixed: reworded to three findings from the update
F11
editorial-advisory
·
(low confidence) "stronger" sign-in means and "first" containment are the entry's characterisation.fixed: takeaway says eID and other sign-in means stayed open and drops first
F11
editorial-advisory
·
(low confidence) The actively-exploited tag and exploited status restate iteration 1.left: advisory; body, summary and sourcing note attribute the claim to CCCS and VulnCheck

Iteration #3 NEEDS_FIXES · 5 findings (truth=2, editorial=1, advisory=2) · Claude Sonnet 5.5 · 15m 01s

F-codeSectionItem · URL/quoteVerifier summaryRemediation · outcome
F3
claim-not-supported
·
(low confidence) The first body paragraph says the vendor-file lines load a router script for iPhones and iPads; Ohey says the 13 November router returns an empty file to anything that is not an iPhonfixed: now says for iPhones
F3
claim-not-supported
·
(low confidence) The update section says Huntress separates the campaign from the SCKit activity; Huntress states the distinction for the crypto-named decoys and says SCKit appears to leverage a thirdfixed: sentence reworded to the decoy-session distinction and the SCKit description
F18
?
·
The 2026-11-12 SCKit entry already tells readers to update every technician host client to 26.6.5 or later, so actions[0] overlaps one of its steps.left: this entry is the canonical patch task for CVE-2026-84869 and the SCKit action carries the step as one of several; the overlap predates this run
F11
editorial-advisory
·
(low confidence) The actively-exploited tag and exploited status restate iterations 1 and 2.left: advisory; the body, summary and sourcing note attribute the claim to CCCS and VulnCheck and CISA KEV absence is stated
F11
editorial-advisory
·
(low confidence) The Symantec and Unit 42 bodies carry file names of legitimate signed executables and loader DLLs.left: advisory; they are behavior anchors for side-loading, not hashes, addresses or attacker domains, and Symantec says the pairing does not by itself identify the actor

Iteration #4 NEEDS_FIXES cap-breach · 2 findings (truth=0, editorial=1, advisory=1) · Claude Sonnet 5.5 · 16m 18s

F-codeSectionItem · URL/quoteVerifier summaryRemediation · outcome
F8
needs-more-research
·
(low confidence) The main analysis had no Exposure and no Detection line although the Huntress source supports both; the detection content sat only in the update section and actions[2].fixed: an Exposure line and a Detection line were added to the main analysis (with the post-patch compromise check), and two uncited or low-value details were cut to keep the analysis under 550 words
F11
editorial-advisory
·
(low confidence) techniques[] was not revisited for the Huntress source: the exploited chain supports T1203 and T1199 among others, and T1211 is not described by any cited source.fixed: T1203, T1199, T1547.001, T1218.007, T1059.001 and T1059.003 added and T1211 removed; the record's fields now name techniques

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

2026-11-17T0355Z-intel · Sonnet 5.5 · window 26 h · 6 entries published

Verification & coverage notes

Window: gap 24.0 h (daily cadence, previous intel fire 2026-11-16T0356Z), window 26 h. No CISA KEV addition fell inside it (work/2026-11-17T0355Z-intel/kev-window.txt, catalogue 2026.11.13) and no covered CVE newly flags ransomware use. The outage-backfill duty did not apply.

Published: six new entries and two update records. New: OptimJack poisoned vendor scripts (high, University of Basel capture), TGR-STA-0049 PerfectPioneer and KUKURAT (notable), Symantec's AI-assisted retailer intrusion (notable, first coverage from the symantec-security-com lookback), the GTIG AI Threat Tracker (annual-report, notable, only the new findings), BundID mass logins (routine, no Swiss portal reported), and Craft CMS CVE-2026-123488 (notable, first coverage from the cccs-alerts and vulncheck lookbacks; exploitation attributed to CCCS and VulnCheck, not asserted). Updates: the Fortinet hardening entry (FG-IR-26-273 now covers FGCP HA clusters; main analysis cut to stay under 550 words) and the ScreenConnect CVE-2026-84869 entry (Huntress confirms exploitation against MSP technicians; actions re-derived). No deep dive: no candidate cleared the bar.

  • Single-source: the Unit 42, Symantec and GTIG entries rest on the vendor holding the telemetry; the Craft CMS exploitation claim rests on CCCS's unnamed open-source reporting plus VulnCheck's KEV flag and is worded as reported. The BundID entry's credential-stuffing label is Handelsblatt's, not the BSI's.
  • borderline-drop: Dassault 3DEXPERIENCE Station Launcher CVE-2026-93664, CVE-2026-19732 and CVE-2026-77033 (unauthenticated RCE rated 10.0, no fix named, no exploitation), no constituency footprint established; doubt about relevance resolves to drop.
  • borderline-drop: NoisyTCP CVE-2026-99288, denial of service only, about 14 hours on a quiet bare-metal host in the researcher's test, no exploit, kernel fix rides the regular cycle.
  • borderline-drop: SharePoint CVE-2026-124223 and CVE-2026-124222 added to the November table on 2026-11-16, authenticated 8.8, Exploitation Less Likely, not exploited or disclosed. Control Web Panel CVE-2026-124020 and CVE-2026-124021, hosting-panel product, no public-sector footprint, no exploitation. Zyxel CPE CVE-2026-79614, WAN access off by default, ISP-customised models excluded, no exploitation. Splunk SVD-2026-1101, advance notice with no vulnerability content until 2026-11-18.
  • borderline-drop: NCSC.ch post 13180 on the six Danfoss AK-SM 800A flaws and its 13054 edit for XWEB EVO, and its N-central rating change to actively exploited, the Danfoss, XWEB EVO and N-able entries already carry these facts, and "vendor firmware updates are available" would read as a fix where Midnight Blue and the BSI say the latest firmware is still affected.
  • borderline-drop: Microsoft 365 outage of 12 to 14 November and the outage-lure phishing campaign (Microsoft Threat Intelligence, 2026-11-16), availability and dependency event with minor Swiss federal effects; the lure is blocked by Defender for Office 365, has no confirmed credential capture and no actor.
  • borderline-drop: the ARTEX reports from Amazon, Cloudflare and Talos (2026-11-13), saturated story with 13 sources in the store, primaries before the 72 h developing window, marginal delta. Barracuda's ARTEX-pattern cloud intrusion (primaries 2026-11-10 to 11-13, out of window, action only for notified customers). Project Zero's "Full X" policy and the Proofpoint ITM macOS flaw (awareness, no technique; no CVE or fix).
  • borderline-drop: Entre Bièvre et Rhône communal breach, Greiz-Ronneburg hospital extortion, Maileva (22 million recipients), Quantum Systems QBase update compromise, the Andropov hoax e-mails, incident floor or out of nexus: no vector, actor or transferable behavior beyond the impact, no Swiss party, or a mechanism that rests on relayed paywalled reporting.
  • borderline-drop: Aargau CyRA pilot, Vaud CYBER26 exercise, CNIL's territorial-authority sanctions (2026-11-09/10) (awareness or outside the window; Cisco's eMudhra advisory v1.12 and Fortinet's Node2Shell advisory revisions) no impacted product, no material delta.
  • Coverage backlog: two rows added, both held on named conditions with expiry 2026-12-01 (TYPO3-PSA-2026-002 pre-announcement, whose release published at 11:00 UTC after this fire began; Oracle's Critical Security Patch Update due 2026-11-17 after this fire). No earlier row was open.
  • Next fire: Oracle's release (pre-release notice of 2026-11-12: 652 patches, Fusion Middleware 53 unauthenticated rows up to CVSS 10.0, E-Business Suite 22 unauthenticated rows up to 9.8) and TYPO3's advisories land after this fire; the 2026-11-18 fire must read the risk matrices and the TYPO3-CORE-SA pages before dropping any row, and Splunk publishes its advisories on 2026-11-18.
  • Verification: four iterations, low-residual early exit on iteration 4 (truth 0, editorial 1, advisory 1, residual 1). Iteration 1 found 11 truth defects, mostly attribution drift in the new entries (OptimJack's capability and dates spread across all four providers, the Ohey citation date, PerfectPioneer's HTTP/2 fallback and blocklist scope, GTIG's stealth-for-velocity sentence) and three in the ScreenConnect update (a count spliced onto another date, a nine-seconds start, an HVCI inference in a pre-existing sentence); all were fixed. Iterations 2 and 3 narrowed to small residuals (Ohey dating, HTTP/2-only variants, the aggregator's BSI attribution of the credential-stuffing label, router scope, an SCKit clause). Iteration 4's editorial finding (the ScreenConnect entry had no Exposure or Detection line and an unrevisited ATT&CK mapping) was fixed after the pass and not re-verified. Left as advisory by design: Craft's actively-exploited tag and exploited status (attributed to CCCS and VulnCheck in the body, CISA KEV absent), the file names of legitimate signed executables in the Symantec and Unit 42 bodies, and the ScreenConnect action that overlaps one step of the SCKit entry's action. The next quality audit should give the ScreenConnect and Fortinet entries an independent pass.
  • Candidate sources not added: Dassault Systèmes advisories listing and Craft CMS GitHub advisories (single-item yield), enea-insights (telecom-only relevance).
  • Source health: 321 records probed; one flagged (anssi-cyber-gouv-actualites, challenge shell), repaired with a sitemap recipe and clean on the re-probe.
  • Process notes: the reader pool is dead (402 on every key, 401 on the anonymous tier) and extract on a 403-walled host ends in a chain of reader errors before it fails; WebFetch's summariser refused an exploit-kit research page, so such pages read through extract.
  • Coverage gaps: sygnia (403 on every transport), belden-psirt (403 on every transport), lab52 (feed reset, WebFetch 503), connectwise-trust-center (bulletin page 403, GitHub Atom feed reads), handelsblatt.com (extract returned a script shell; a syndicated copy was read), inside-it-ch Plus articles (paywall), the Splunk and Oracle releases (not yet published).
  • Watchlist: products checked=0, hits=0; suppliers checked=0, hits=0
  • Essential-coverage: all 21 essential records attempted.

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