Confidential · 12 Sep 2026
🛡️ Daily Cybersecurity Briefing — 2026-09-12 (Saturday)¶
Window: last 24–48h (Sep 11–12). Severity: 🔴 CRITICAL · 🟡 HIGH · 🟢 MEDIUM.
Threat level GUARDEDVictims L30D 72Top actor QilinM&A L30D $94.5M
Number of the day151 million — the AI-query exchanges Anthropic attributes to a single distillation campaign it says was run by operators linked to Alibaba, the largest such attack the company has ever measured, disclosed in a Sept 10 threat-intelligence report naming six other China-based labs. Anthropic
💼 M&A ACTIVITY¶
No new cybersecurity M&A deals announced in the Sep 11–12 window.MediumWeekend quiet.
L30D summary (Aug 13 – Sep 12): 5 deals tracked, $669.5M+ in disclosed value. Biggest: Munich Re (via HSB) → At-Bay $575M (cyber insurtech/MDR); Datavault AI → CyberCatch Holdings $94.5M all-cash (AI-enabled compliance). Consolidation theme unchanged: insurers buying prevention capability rather than just underwriting risk (Munich Re/At-Bay), and platforms bolting on AI-security or agentic-AI tooling at undisclosed values (Fortinet/Virtue AI, Brinqa/PlexTrac, Socure/Fravity). The prior 30-day window's two smallest rounds (A Security $37M Series B, Tenet Security $6M seed, both Aug 12) have now rolled off the trailing window. SecurityWeek M&A Tracker
⚠️ CRITICAL BREACHES & INCIDENTS¶
Florida's DMV confirms the ShinyHunters intrusion this desk flagged as an unverified claim Sep 8 — and the root cause is a single officer's personal device.HighGovernment & Public SectorFLHSMV says it learned of the breach Sep 4, traced entry to a Plant City Police Department user account whose DMV/DAVID-database credentials were improperly stored on the officer's personal device rather than an agency-issued one, and calls the intrusion "quickly mitigated" with no further breach ongoing. ShinyHunters' own claim of 200,000+ driver records remains the attacker's figure, unconfirmed by the state; the intrusion vector (one non-agency device holding law-enforcement-grade database access) is the new, confirmed detail. The Record
General Santos Doctors Hospital, a 280-bed tertiary hospital in the Philippines, listed on Rhysida's leak site Sep 10.HighHealthcare & Life Sciences🟥 Unverified DLS claim — roughly 3.5M files (2.44TB) allegedly exfiltrated, including name-tagged diagnostic scans, cancer-center records with national health-insurance IDs, and a staff register with professional license numbers; no hospital confirmation yet, verify before treating as a breach. Ransomware.live
Spain's national meteorological agency (AEMET) listed by Panzer, a RaaS brand that launched its leak site Aug 5 and already claims 16–21 victims across 11 countries.MediumGovernment & Public Sector🟥 Unverified DLS claim — roughly 5GB allegedly exfiltrated, 20–21 day publication deadline; no agency confirmation yet, verify before treating as a breach. EscudoDigital
🔓 CRITICAL VULNERABILITIES¶
CISA adds a maximum-severity GitLab path-traversal flaw to KEV Sep 11, one day after attackers began exploiting it.CriticalTechnology & SoftwareCVE-2026-85706 (CVSS 10.0) lets an unauthenticated attacker abuse GitLab's repository commits API to read arbitrary files off a self-managed CE/EE server — configs, secrets, anything the app can reach — with no account or user interaction required. Affects versions 18.7–19.1.7, 19.2–19.2.5, and 19.3–19.3.1; patch to 19.1.8/19.2.6/19.3.2 or later. Federal remediation due Sep 14. BleepingComputer · CISA KEV
Check Point discloses two 9.8-rated, unauthenticated RCE flaws in VPN certificate handling across its Quantum Security Gateway line, Spark Firewalls, and Security Management Server.HighTechnology & SoftwareCVE-2026-85102 is a certificate trust-validation bypass during VPN negotiation; CVE-2026-85103 is a heap buffer overflow in certificate decoding — both remote, unauthenticated, no user interaction. Check Point says it has seen no exploitation as of disclosure (Sep 9); LivePatch Take 24 and Jumbo Hotfix Accumulator updates are available now. Given WatchGuard's Firebox flaw took nine months to reach ransomware use after KEV listing, "not yet exploited" is not a reason to delay patching VPN gateways. The Hacker News
🚨 INTELLIGENCE AGENCY ALERTS & POLICY¶
No new CISA/FBI/NSA/NCSC advisory in the Sep 11–12 windowMediumtoday's operative federal action is the GitLab KEV addition above, issued by CISA Sep 11.
🌐 THREAT ACTOR & CAMPAIGN ACTIVITY¶
Baseline DLS volume Sep 11–12: roughly a dozen new named victims across tracked leak sitesMedium, including Safepay (compunnel.com, US IT/staffing), RansomHouse (California School Employees Association), Akira (Eagle Construction, US), Beast (M800/CINNOX, Hong Kong telecom-API provider), and EMPERADOR (EJ Easy Job, Colombia) — none individually clears the bar for a new tracker entry beyond Rhysida/General Santos and Panzer/AEMET, noted above.
Anthropic's fourth threat-intelligence report (published Sep 10, covering activity Anthropic identified and shut down between December 2025 and August 2026) discloses it dismantled five illicit-distillation campaigns from seven China-based AI labs — Alibaba, DeepSeek, Moonshot AI, MiniMax, StepFun, Z.AI, and one more — that generated nearly 190 million Claude exchanges.HighAlibaba's campaign alone accounted for 151M+ exchanges from over 3,500 accounts Anthropic describes as fraudulent, peaking near 3 million queries/day, allegedly to improve its Qwen models. This is Anthropic's own confirmation of the pattern the NSA/CISA/FBI joint advisory AA26-251A attributed to the same six labs by name on Sep 8 — new here is the operator-level detail (Alibaba specifically, exchange volumes, account counts) and that Anthropic frames it as a nation-state-scale IP-extraction operation, not merely policy-violating API use. Anthropic
🌍 GEOPOLITICS¶
Analyst lens: how this week's cyber activity maps to state strategy. Defense · cyber · economics.
Anthropic naming a specific Chinese company (Alibaba) and a specific number (151 million exchanges) as the source of the largest AI-capability extraction it has measured turns a diplomatic-hedge issue into a quantified, single-vendor accusation three days before the CISA/FBI/NSA advisory's own six-lab attribution had fully settled into coverage — and it lands two weeks ahead of the Sep 24 Trump-Xi summit.HighTechnology & SoftwareCybersecurityWhere AA26-251A (Sep 8) named six labs generically as running "industrial-scale" distillation, Anthropic's own report puts a dollar-and-scale figure behind one company's alleged conduct, which is harder for Beijing to wave off as generic state-linked activity and harder for US trade negotiators to leave out of the agenda. Watch whether Alibaba or the Chinese government issues a direct rebuttal (as opposed to the usual boilerplate denial), and whether this becomes a specific line item in pre-summit talking points rather than background noise. Anthropic
A state government's most sensitive law-enforcement database access failing because one officer stored credentials on a personal phone is a policy problem no patch fixes, and it is happening in the same month a private identity-verification vendor (IDScan.net) leaked 153M+ driver's licenses for unrelated reasons.HighGovernment & Public SectorTechnology & SoftwareFlorida's DMV breach and IDScan.net's slow-walked disclosure are two separate incidents with one shared structural cause: identity-document infrastructure — public and private — runs on access-control assumptions (agency-issued devices, indexed disclosure) that keep failing in ordinary, boring ways rather than exotic ones. For portfolio companies serving government identity/DMV contracts, the audit question is device-issuance policy enforcement, not just encryption-at-rest. The Record
GitLab's flaw going from disclosure to confirmed exploitation in under 24 hours, set against WatchGuard's nine-month gap reported here Sep 11, brackets the actual range of the "known exploited" clock rather than picking one end of it as typical.MediumTechnology & SoftwareBoth are now federally mandated remediation items; the operational lesson for portfolio companies running self-managed developer infrastructure (GitLab, Jenkins, GitHub Enterprise) is that internet-facing dev-tooling now sits in the same rapid-exploitation tier as edge network appliances, not a slower "internal tooling" risk class. BleepingComputer
Two previously-unseen ransomware brands (Vexy, first observed Sep 10; Panzer, live since Aug 5) each reaching double-digit, multi-country victim counts within roughly a month of appearing is a market-structure signal worth tracking rather than dismissing as routine churn.MediumCybersecurityIf barriers to standing up a credible RaaS operation — leaked builders, commoditized affiliate recruitment — keep falling, the leaderboard's top ranks matter less than the total addressable pool of active brands at any given time; this desk has added the pattern to the signals watchlist. Ransomware.live
M&A activity
Socure → Fravity—
Brinqa → PlexTrac—
Munich Re (via HSB) → $575M—
Fortinet → Virtue AI—