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

2026-11-16T0356Z-intel

One pipeline fire, in full · intel run of 2026-11-16 · 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/2026-11-16/2026-11-16T0356Z-intel.md.

Run telemetry

2026-11-16T0356Z-intel intel prompt v4.25 publish ok
1h 04m duration 0 published 1 updates
Claude Sonnet 5.5 (claude-sonnet-5-5) main agent
S1 Claude Sonnet 5.5 (claude-sonnet-5-5)
Items returned
4
Duration
22m 49s
Tool calls
6 WebFetch13 WebSearch95 bridge
Cited sources
0 of 28 in slice
S2 Claude Sonnet 5.5 (claude-sonnet-5-5)
Items returned
5
Duration
25m 41s
Tool calls
2 WebFetch28 WebSearch120 bridge
Cited sources
0 of 19 in slice
S3 Claude Sonnet 5.5 (claude-sonnet-5-5)
Items returned
7
Duration
23m 07s
Tool calls
2 WebFetch9 WebSearch100 bridge
Cited sources
0 of 16 in slice
S4 Claude Sonnet 5.5 (claude-sonnet-5-5)
Items returned
10
Duration
23m 39s
Tool calls
3 WebFetch24 WebSearch70 bridge
Cited sources
1 of 12 in slice

Verification

✓ double-CLEAN · Sonnet 5.5 ×2 #1 NEEDS_FIXES · Sonnet 5.5 · t=1 e=2 a=1 #2 CLEAN · Sonnet 5.5 · t=0 e=0 a=3 #3 CLEAN · Sonnet 5.5 · t=0 e=0 a=4

Deep dive

·

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 · 2 notes · 1 status · 1 max_staleness_days · 1 consecutive_fetch_failures · 1 last_successful_fetch · 1 consecutive_quiet_periods.

SourceChangeFrom → ToReason
cis-msisac-advisoriesadded candidate· → ·public-sector multi-vendor advisory feed that reads direct with 15 dated items and gives a cross-vendor release calendar for the S1 scheduled-release duty
bka-presseportaladded candidate· → ·German federal police primary for Operation Endgame and other takedowns; carried a fact (StealC rebuild) that Europol lacks; feed reads direct
arctic-wolf-labsstatuscandidate → activepromotion_due in the state digest: cited by published entries from 5 distinct runs
edoeb-ch-pressmax_staleness_days· → ·health flagged the listing stale at 61 days; the regulator releases a few items per quarter and the listing is a stale cache, the sitemap recipe finds the real releases; limit set to 120
safeonweb-benotes· → ·health still classes a shell; the whole host answers HTTP 500 on every path and language, a server-side outage with nothing to repair; kept active, a 5xx never demotes
ico-uk, europol-newsroom, tio-ch-attacco-informatico, nict-cynex-lettice, le-monde-info, vd-grand-conseil, connectwise-trust-center, darkreading, github-advisory, msrc-update-guide, helpnetsecuritynotes· → ·recipe corrections and confirmations found by the sub-agents (Welsh ICO mirror, Europol dates, tio.ch topic page, NICT JSON list, Le Monde Next.js paths, Vaud detail links, CNA commits API, Dark Reading extract falling to the reader, GitHub advisories API, MSRC OData filter, Help Net Security hrefs)
belden-psirt, sygnia, ico-uk, lab52consecutive_fetch_failures· → ·transport failures on every route (no demotion: 403 or anti-bot blocks never demote)
bleepingcomputerlast_successful_fetch· → ·fetched and used: the FBI FLASH on ARTEX reached the run through its feed
(71 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:extract → webfetch → bridge:jina403 transport-403
403 on every transport; the reader pool answered 402 on every key
none; candidate record, Belden products already appear in earlier PHP2Shell entries
sygniahttps://www.sygnia.co/blog/bridge:extract → webfetch → websearch403 transport-403
direct 403 on extract and WebFetch, no trafilatura body, reader pool exhausted
none; the record is pinned fetch_method blocked and earlier fires failed the same way
lab52https://lab52.io/blog/feed/bridge:feed → webfetch → bridge:extractn/a transport-timeout
feed connection reset, WebFetch HTTP 503; extract returned six undated cards through the reader with no hrefs
none; no in-window Lab52 item could be identified
ico-ukhttps://ico.org.uk/about-the-ico/media-centre/news-and-blogs/bridge:ico-uk.enforcement → bridge:extract → webfetch → websearch403 transport-403
upstream HTTP 403 on the sitemap, extract and WebFetch; reader pool exhausted
WebSearch found no ICO action dated in the window; S2 found the Welsh mirror cy.ico.org.uk unblocked, recorded as a recipe note

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 pdf ×1
  • tools/fetch_source.py feed ×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 · 4 findings (truth=1, editorial=2, advisory=1) · Claude Sonnet 5.5 · 6m 43s

F-codeSectionItem · URL/quoteVerifier summaryRemediation · outcome
F9
surface-contradiction
·
The new section relays the FBI advice to keep ConnectWise connections disabled until ConnectWise gives restoration instructions, although ConnectWise had already published them on 2026-10-28.fixed: the clause was removed; the section keeps the FBI review-by-application-ID request and its advice not to delete the enterprise application before evidence is preserved
F8
needs-more-research
·
The section drops the FLASH hardening levers (restrict Kudu/SCM, disable basic publishing authentication, inventory peripheral internet-facing applications), so the newest section gives no decision fofixed: the final paragraph now carries the FBI recommendations; the main analysis, already over its length bound, was left unedited
F14
?
·
(low confidence) "ten sectors" in the headline, summary and record summary is the pipeline count of a list the FLASH introduces with "including"; the FLASH states no number.fixed: headline says many sectors, summary and record summary name sectors without a count; the run record no longer says ten
F11
editorial-advisory
·
T1059.001 is absent from techniques[] although the section describes PowerShell enumeration, and the XXE metadata query is grouped under internal networks while the FLASH lists it under privilege escafixed: T1059.001 added and the sentence regrouped as privilege escalation and internal discovery

Iteration #2 CLEAN · 3 findings (truth=0, editorial=0, advisory=3) · Claude Sonnet 5.5 · 6m 35s

F-codeSectionItem · URL/quoteVerifier summaryRemediation · outcome
F11
editorial-advisory
·
The frontmatter summary is six long sentences against the documented 1 to 3 sentences; each update appended one.left: advisory; accurate, the audit may tighten the summary
F11
editorial-advisory
·
(low confidence) The FLASH collection row lists T1074.002 and the section describes new 100 GB volumes, but techniques[] has no T1074 id; the FLASH does not tie the id to a bullet.left: advisory; no source ties the id to the described behaviour
F11
editorial-advisory
·
The record summary omits the ConnectWise-tenant review ask that the section states; not an overstatement.left: advisory

Iteration #3 CLEAN · 4 findings (truth=0, editorial=0, advisory=4) · Claude Sonnet 5.5 · 4m 49s

F-codeSectionItem · URL/quoteVerifier summaryRemediation · outcome
F11
editorial-advisory
·
The frontmatter summary is about 243 words in five long sentences against the 1 to 3 sentence contract (carried over from iteration 2).left: advisory
F11
editorial-advisory
·
(low confidence) The Exposure and Defender takeaway lines still read identity-only while the newest section carries non-identity levers from the FBI; the takeaway is attributed to Expel and not disproleft: advisory; the main analysis already exceeds its length bound and the new levers sit in the section
F11
editorial-advisory
·
(low confidence) "Telemetry the FBI lists" includes IAM DryRun-flag tests, which the FLASH puts in its observed-activity row rather than its indicator lists; techniques[] has no T1074.002.left: advisory; the item is the FBI observed behaviour and the lead-in sits within the FLASH content
F11
editorial-advisory
·
The sourcing_note is now about ten sentences against the two-sentence provenance standard; accurate.left: advisory; the audit may tighten it

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

2026-11-16T0356Z-intel · Sonnet 5.5 · window 26 h · 0 entries published

Verification & coverage notes

Window: gap 24.0 h (daily cadence, previous intel fire 2026-11-15T0356Z), window 26 h. A Sunday-to-Monday window: the newest item on every essential record is dated 2026-11-13 or earlier, no CISA KEV addition fell inside it (work/2026-11-16T0356Z-intel/kev-window.txt, catalogue 2026.11.13) and no covered CVE changed its ransomware flag. The outage-backfill duty did not apply (normal cadence).

Published: no new entry and one update record on 2026-10-22/artex-storm-0940-agentic-llm-pentest-spraying-oauth-abuse. The FBI's FLASH-2026-1113-001 (2026-11-13) gives a US authority's account of a variety of actors using ARTEX since at least August against sectors including government, and lists the observed steps (cloud-trust and B2B-federation pivots, Kudu abuse, Active Directory credential theft, exfiltration preparation) and the telemetry to check, none of which the entry carried. It cleared PD-11(d) as a development on covered ground; priority stays notable. The do-now ConnectWise application actions already sit on 2026-10-18/connectwise-cloud-backup-skykick-m365-app-credential-abuse, so actions[] did not change. Slice-completion check: every slice record of all four domains has a ledger row; no continuation was needed. state/coverage_backlog.md had no open row.

  • Single-source: none published as a new entry. The update rests on the FBI FLASH alone for its table; the entry's sourcing note says the FLASH consolidates the vendor reports it cites.
  • borderline-drop: PostgreSQL 18.7, 17.12, 16.16, 15.20 and 14.25 (CVE-2026-6460 and 51 more), dropped on 2026-11-14 for "no public PoC"; the finder's write-up of 2026-11-09 now shows the crash step and a memory-disclosure step but not the code-execution step, the flaw needs a role that can create objects, no source reports exploitation and the fix is a scheduled minor release inside the regular cycle. Drop stands; the audit may revisit if exploitation or the missing step surfaces.
  • borderline-drop: Cyrus SASL SRP authentication bypass CVE-2026-129483; the SRP plugin is built only with --enable-srp (default off), no source names a product that ships it, no exploitation; upstream removed the plugin on master and no release has followed 2.1.28. Apache Shiro CVE-2026-129349, needs the non-default authenticationCachingEnabled on an LDAP realm and a known username, no exploitation.
  • borderline-drop: N-able N-central Hotfix 3 postponed (heise 2026-11-15, returned by S1 and S4); the entry already says a follow-up hotfix is planned and the stopgap is unchanged; N-able's own status page is a script-rendered shell no transport reads.
  • borderline-drop: Operation Endgame round of 2026-11-12 (Gootloader, CastleRAT, StealC; Europol and BKA), no Swiss participant and no technique content; the victim-notification channels are generic hygiene; also dropped on 2026-11-15.
  • borderline-drop: Proofpoint "Europe's Threat Landscape in 2026" and ENISA's 2026 State of Cybersecurity in the Union (both flagged ANNUAL REPORT); Proofpoint is trend statistics over behaviours the store already covers (device-code phishing, adversary-in-the-middle kits, webmail exploitation without a CVE for TA488) and its one Swiss datum is TA2730 lures aimed at brokerage customers; ENISA is a policy-level EU report outside Swiss jurisdiction with no technique content.
  • borderline-drop: Symantec's AI-assisted intrusions (US retailer; Orion RabbitMQ Erlang-cookie lateral movement); Symantec states the techniques are not new, one US victim, nothing a Swiss SOC would hunt differently; the AI thread is carried by the ARTEX, CyberStrikeAI and 3AM entries. Operation Snowball (Malbear Labs; Cisco BroadWorks XXE CVE-2026-20320, KEV 2026-11-10), Unit 42's CL-STA-1238 (Cairn) and Enea's SilentAir and GTPDoor, an unnamed European telecom with no Swiss customer, US-centred targets and Asian operators.
  • borderline-drop: KEV rows of 2026-11-09 to 2026-11-13 that no entry covers, Cisco BroadWorks CVE-2026-20320, GPUStack CVE-2026-102895, Openwave Webtop CVE-2026-111793, UnrealIRCd CVE-2026-114713. The earlier dispositions stand. Cisco's advisory v1.2 of 2026-11-13 says it has no direct evidence of active exploitation of BroadWorks; UnrealIRCd's maintainer now confirms exploitation (The Hacker News 2026-11-14, heise 2026-11-15), but it needs an IRC network with two or more linked servers and no constituency footprint appears.
  • borderline-drop: Quantum Systems QBase update-server compromise (sources of 2026-11-12 and 2026-11-13, no Swiss customer named, reports conflict on whether Bundeswehr systems received the file); Eichstätt municipal ransomware (no vendor or actor named, letters to residents the only new fact); Maileva, 22 million recipients (no actor, account misuse only, no Swiss client); Amadeus booking-display breach (no Swiss victim, root cause disputed); Fireworks AI incidents (US downstream victims, origin already in the Cursor entry); BundID credential stuffing (one community blog, no operator or press confirmation); Berlin Rhysida leak interview (one analyst's assessment, no new fact on the incident); outage-themed Microsoft 365 phishing (single report relaying a vendor social post, commodity lure, US observations, no confirmed compromise).
  • out-of-window: Saxony emergency-services pager exposure, heise 2026-11-12 and Spiegel 2026-11-11, window_hours=26; the 2026-11-13 verifier also removed it as below the incident floor and the 2026-11-15 fire dropped it.
  • borderline-drop: SAP Patch Day critical rows beyond the four CVSS 10.0 rows (authenticated or CSRF), Fortinet FG-IR-26-193 (CVSS 7.7, man-in-the-middle precondition), GitHub advisory database niche criticals, the Cisco ISE row, Vaud Grand Conseil interpellation 26_INT_155 (a member's questions, no finding), React Server Components CVE-2026-126187 (no exploitation or proof of concept; dropped on 2026-11-12, 2026-11-13 and 2026-11-15).
  • Next fire: Oracle's Critical Security Patch Update is due 2026-11-17 (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, Analytics up to 10.0). The 2026-11-17 and 2026-11-18 fires must read the release's risk matrices before dropping any row.
  • Candidate sources not added (the guide is about one per run; two added): darktrace-blog (feed lags the site), malbear-labs (no feed, first appearance), enea-insights (telecom-only relevance).
  • Source health: 320 records probed; two flagged. edoeb-ch-press is fixed (staleness limit set to its quarterly cadence). safeonweb-be stays flagged: the whole host answers HTTP 500, a server-side outage with nothing to repair, documented in its notes.
  • Process note: the reader pool is dead (402 on every key, 401 on the anonymous tier), so extract on a 403-walled host ends in a long chain of reader errors before it fails; the recipe notes record the mirrors and APIs that work instead.
  • Verification: three iterations, confirmed CLEAN (iterations 2 and 3, both independent cold passes, no truth or editorial finding). Iteration 1 found three real defects in the new section and fixed them: a superseded FBI instruction about ConnectWise restoration was dropped, the FLASH's own hardening levers were added, and the sector count was removed because the FLASH states none; its advisory (T1059.001, metadata-query grouping) was applied. Advisories left as they stand: the entry's summary and sourcing note are longer than the 1-3 and two-sentence standards after six records (the audit may tighten them), and T1074.002 is not mapped because the FLASH ties it to no named behaviour.
  • Coverage gaps: ico-uk (403 on every rung, Welsh mirror readable); sygnia (blocked on every transport); belden-psirt (403 on every transport); lab52 (feed reset, WebFetch 503, extract returned undated cards); safeonweb-be (host outage, HTTP 500); connectwise-trust-center (403, CNA commits API readable); databreaches-net (feed readable, article pages 403); previdian (listing and feed not sweepable).

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