2026-09-17T0409Z-intel

Verification & coverage notes

Standard 26-hour window (24 hours since the previous fire). No closed-source drops this window. No product or supplier watchlist is configured for this deployment, so both sweeps are no-ops (Watchlist: products checked=0, hits=0; suppliers checked=0, hits=0).

Every CISA KEV addition since the previous fire got a disposition:

• CVE-2026-58704 (Google Pixel modem, zero-click privilege escalation) → published: 2026-09-17/cve-2026-58704-google-pixel-modem-zero-click-eop (priority: high).
• CVE-2026-76460 / CVE-2026-76423 (Cisco ISE auth bypass) → published: 2026-09-17/cve-2026-76460-cisco-ise-auth-bypass-root-rce (priority: critical, newly disclosed, confirmed actively exploited, no workaround beyond ACLs).
• borderline-drop: CVE-2026-87886 (Acronis Backup plugin for cPanel & WHM / Plesk extension) — KEV-listed and confirmed exploited, but the attack surface (Linux shared-hosting backup tooling requiring an existing low-privilege tenant account) is narrow and not confirmed to run on any platform a Swiss federal/cantonal/communal administration would itself operate; relevance is indirect (smaller hosting providers hosting public-sector sites).

New entries (7):

1. cve-2026-76460-cisco-ise-auth-bypass-root-rce (critical, vulnerability), unauthenticated Cisco ISE API auth bypass, confirmed exploited, found via a customer support case.
2. cve-2026-58704-google-pixel-modem-zero-click-eop (high, vulnerability), Pixel modem zero-click privilege escalation, CISA KEV, exploited in limited targeted attacks per Google/TechCrunch.
3. kairos-libercourt-commune-ransomware-confirmed (notable, incident); a small French commune confirms a ransomware attack and data theft, days after the extortion actor Kairos claimed it on its leak site; out-of-nexus, included for the target-profile pattern this actor has now shown twice against small municipal administrations.
4. aepd-first-ai-agent-breach-notification (notable, policy), Spain's data protection authority discloses the first GDPR breach notification attributed to an autonomous AI agent. aepd.es was added to tools/check_run.py's national-authority carve-out list this run, matching the existing precedent for France's CNIL.
5. ddrop-dram-interposer-defeats-confidential-computing (notable, research), a DDR5 hardware interposer defeats Intel TDX/SGX and AMD SEV-SNP confidential-computing guarantees; no CVE, no vendor fix (a physical-access class of attack).
6. phantomraven-npm-llm-generated-infostealer (notable, threat), CrowdStrike attributes an LLM-generated npm infostealer to a self-described bug-bounty hunter.
7. mandiant-ai-risk-resilience-report-2026 (high, annual-report, deep dive); Mandiant's second annual AI Risk and Resilience report; eight case studies of AI-agent abuse in real intrusions and red-team engagements. No deep dive had published yet today, so this is the day's single deep dive.

The AEPD story was found independently from two directions (the home-region/sector track and the research track), tracing to the same primary, the authority's own blog post, picked up by heise online. It was composed once, citing both, rather than as two entries.

Coverage-backlog re-checks this run, all "no change" except one resolution:

• Kairos/Ville de Libercourt, resolved, published as entry 3 above.
• ShinyHunters/Kimberly-Clark, no change (leak-site/aggregator only; the actor's own negotiation deadline passed 2026-09-16 with still no victim statement or reputable independent journalism).
• inside-it.ch's Insel Gruppe article, still blocked (a persistent "Security Checkpoint" HTTP 429 on every transport tried, an 18th+ consecutive fire).
• TheGentlemen/Ixa Systems SA, Krybit/UICC, ShinyHunters/Medela AG, SafePay/reichenau.at, no change (leak-site-tracker-only, no victim statement, no reputable independent journalism).
• NovoCure, no new Swiss public-sector nexus found; not republished.
• The Siemens S7 PLC advisory and three lower-priority research items held in the backlog (AWS credential-stuffing, an Exodus wallet installer RAT, a JSCeal deobfuscation pipeline), checked, no material development; remain below the bar for their own entry.
• Ville du Tampon and Familea (French municipal incidents), re-checked, still no named mechanism, actor or data-theft claim from any party.

New in-window lead, held for further corroboration (not published): AFPA (France's national vocational-training agency), a third data-theft claim in a month, from two named criminal handles against two distinct claimed datasets, one with a claimed access-control mechanism, examined by two independent breach trackers. Still no statement from AFPA and no reputable independent journalism found despite a targeted search, so it does not yet clear the bar for inclusion. Logged for re-check on a later fire.

Verification: six iterations. The first four each found genuine truth or editorial defects (citation mismatches, an inflated advisory count, an uncited and overstated comparison claim, ATT&CK mapping precision) and were remediated in place. The fifth returned clean on everything but one minor ATT&CK-completeness point; the sixth, an independent cold-confirmation pass, found one further missing citation and a handful of additional ATT&CK ids a vendor's own mapping table names, both fixed. With the sixth iteration's residual at truth 0 / editorial 1, both fixed, this closes the loop without a further confirmation pass. One acknowledged residual: four low-support ATT&CK ids from CrowdStrike's PhantomRaven mapping table (domain registration, the scripting-interpreter execution technique, system locale discovery, automated collection) were judged too granular relative to that entry's own descriptive depth to add without overstating precision, and are left unmapped by deliberate choice rather than oversight.

Sources: changes listed above; the tp-link-omada-psirt listing page remains unreachable (a candidate replacement URL was checked and rejected this run, it returns only a cookie-consent shell, not an advisory listing).