2026-06-05-2c6574c4

• Dropped — CVE-2026-41283 (OpenStack Mistral): a sub-agent surfaced this as an "unauthenticated, CVSS 9.9, pre-auth RCE." A Phase 2 spot-check of the primary advisory (OSSA-2026-020 via oss-security, 2026-06-03) contradicts that framing: the policy-enforcement bypass requires any authenticated user, the advisory carries no CVSS score, and there is no ITW exploitation. The "unauthenticated 9.9" claim traced to a low-reliability aggregator (thehackerwire.com) and did not survive verification. Authenticated, no-PoC, no-ITW, CVSS-unconfirmed → does not clear a § 2 inclusion gate.
• Dropped — CVE-2022-0492 (Linux cgroups v1 release_agent): already covered as the 2026-06-03 deep dive (the KEV re-entry was that day's story). The only new element offered was the CISA KEV remediation deadline (5 June), which is a US FCEB compliance date with no jurisdictional weight in CH/EU and is not material new development (PD-13). Excluded.
• Dropped — CERT-FR weekly bulletin CERTFR-2026-ACT-024: a national-authority consolidation of CVEs already covered individually (Samba CVE-2026-4408/-4480, PAN-OS CVE-2026-0257); no new in-window delta. Roll-up digests are not cited in place of the primaries they summarise (PD-12).
• Dropped (editorial relevance, off-audience or low operational signal): ECB "dear CEO letter" on AI-cyber risk (supervisory/policy framing, thin for a technical SOC, weak corroboration); Europol Operation KRATOS 2 (illegal-streaming takedown) and the Spain fake-EU-document-factory takedown (law-enforcement, little defender action); IMA Diligence Services breach (US-only insurance, limited CH/EU nexus); Luna Moth / Weil Gotshal extortion (single reported ransom figure, low novelty); Hola Browser XMRig-miner supply-chain incident (consumer browser, already remediated, ~0.1% impact).
• Single-source items (named): VerdantBamboo / UNC5221 (§ 1) — Volexity only, HIGH-reliability primary IR research, marked [SINGLE-SOURCE] in-line. CVE-2026-34906 / -34907 (§ 2) — CERT Polska only, national-CERT carve-out as primary disclosing party (PD-5).
• Reduced confidence — only aggregator sources: TA4922 (§ 1) — Proofpoint is the primary research origin but its blog is a JS-rendered SPA that could not be fetched directly; coverage rests on The Hacker News and BleepingComputer reporting Proofpoint's findings, both news-aggregator hosts.
• No item met the Immediate Actions bar this run (no confirmed active mass-exploitation requiring same-hour action), so the § 0 callout is omitted.
• Noted but not pursued: a JFrog report on IronWorm (a Rust-based worm in the Shai-Hulud / TeamPCP npm supply-chain cluster) surfaced via a newsletter digest. The Shai-Hulud cluster is a long-running campaign already consolidated in the 2026-W22 weekly, so it is deferred under the long-running-campaign rule (PD-8) rather than re-opened here.
• Verification: 3 iterations (all on the Sonnet cti-verification-alt variant — the Opus cti-verification spawn was blocked twice by Anthropic's violative-cyber-content classifier while ingesting the brief; the Sonnet variant was used to satisfy the mandatory verification loop). Iter 1 → NEEDS_FIXES (SecurityWeek mis-dating/mis-attribution; an unsupported conference claim; a broken victim-statement URL); iter 2 → NEEDS_FIXES (Xint Code mis-attributed to Wiz vs. Theori; an over-counted CVE total; a Salesforce vector projected onto DentaQuest); iter 3 → CLEAN. Residual: 0.
• Coverage gaps: inside-it-ch (persistent 403, no usable Wayback snapshot — 6+ runs); databreaches-net (403, no Wayback — 6+ runs); sophos-xops (HTTP 503 — 5+ runs); sec-disclosures-edgar (bridge HTTP 500; EDGAR full-text fallback returned 0 Item 1.05 filings in window); proofpoint-blog (JS-SPA, body not fetchable); ncsc-ch-security-hub, safeonweb-be, ncsc-ie, enisa, mandiant-gtig — not fetched or no in-window items this run.

Unmatched action items (migrated)

• Patch Redis to 7.2.14 / 7.4.9 / 8.2.6 / 8.4.3 / 8.6.3 (§ 5). Where patching lags: require a password, bind off the public internet, and apply ACL least-privilege (deny CONFIG and @scripting to application users) — this breaks the public exploit chain. First inventory every reachable Redis instance, especially passwordless ones.
• Update claude-code-action to v1.0.94+ and audit AI-agent CI/CD workflows (§ 3). Review every issues / pull_request_target-triggered workflow that grants an AI agent write scope; never set allowed_non_write_users to "*".
• Hunt for the VerdantBamboo edge-device pattern (§ 1): M365 sign-ins originating from NAS / storage-sync / firewall egress IPs; SSL-VPN re-enablement and admin authentication to perimeter appliances. Enforce MFA on all firewall management/SSL-VPN interfaces and treat MSP access to your perimeter as privileged-insider access.
• Polish public-university operators: shield Wirtualna Uczelnia now (§ 2) — restrict the redirectToUrl endpoint to internal/authenticated sources and hunt access logs for template metacharacters until Simple SA ships a fix (no patch at disclosure).
• Hunt TA4922 tradecraft (§ 1): DLL side-loading where AnyDesk/SyncFuture load from unexpected paths; Python processes touching Chrome/DPAPI credential stores; unsolicited LINE/WhatsApp/Teams contact that pivots to a document.
• macOS fleets: behavioural detection for FlutterShell (§ 1) — Gatekeeper/notarization will not catch it. Alert on apps instantiating WKWebView with a JS message handler that spawns shells, non-browser writes to Chrome's Secure Preferences, and "productivity" apps reaching CDN-fronted infrastructure.
• Patch and watch internet-facing cPanel (§ 1): pin to current release, disable unused modules, alert on admin lockouts and anomalous file-manager/FTP changes.

Migrated from briefs/2026-06-05.md (v2).