Daily brief

Top story

IDScan.net confirmed a breach after 153M+ US and Canadian driver's license scans surfaced on the "Nexus" marketplace — a trove covering 170M+ people, including 10M+ ID cards, 3M+ travel documents, 579,000 medical cards, and 1.1M+ Canadian records, possibly exfiltrated continuously for over a year. The Louisiana identity-verification firm, whose cloud-stored customer data was accessed around September 1, serves car rental companies, retailers, and cannabis dispensaries.

Exploitation & threats

• Wiz observed attackers chaining JFrog Artifactory CVE-2026-42018 and CVE-2026-42016 since August 15 for admin control on self-hosted servers, installing Groovy plugins and Rust backdoors; critical auth-bypass CVE-2026-82329 was mass-exploited separately, and CISA added the Artifactory bugs plus ScreenConnect CVE-2026-84869 to KEV.
• Three clusters — including Qilin operators — exploit Cisco FMC flaws CVE-2026-20079 and CVE-2026-20316: UAT-12197 (JSP web shells, credential theft), UAT-11823 (Sandworm-overlapping tooling, Cyclops Blink), and UAT-11988 (Qilin ransomware).
• Hunt.io ties the July 17 attack on the Borough Council of King's Lynn and West Norfolk to SonicWall SMA1000 CVE-2026-15409 (CVSS 10.0) — a scanner adapted from Rapid7's PoC stole LDAP credentials for 534 Active Directory accounts.
• GitLab CVE-2026-85706 entered KEV with a September 14 deadline after WatchTowr saw in-the-wild probes within a day of disclosure.
• GreyNoise says AI agents on OpenAI's Codex with a DeepSeek model exploited PaperCut CVE-2026-81578/CVE-2026-82078 for a likely Russian-speaking actor, compromising 440 instances across 395 organizations in 48 countries, with domain admin within six hours of first RCE.
• Microsoft ties passkey-themed helpdesk vishing to Storm-3121 (ShinyHunters-linked) and Storm-3032 (Helix), compromising M365 accounts since May via AiTM pages and device-code phishing.
• Anthropic disrupted Midnight Blizzard campaigns in which Claude autonomously rebuilt malware until it evaded detection, targeting 20+ government, defense, and think-tank organizations.

Patch priorities

• GitLab CVE-2026-85706 (CVSS 10.0, unauthenticated arbitrary file read): upgrade self-managed CE/EE to 19.1.8, 19.2.6, or 19.3.2 before September 14.
• JFrog Artifactory: patch CVE-2026-42016, CVE-2026-42018, and CVE-2026-82329, then audit for rogue admin accounts, Groovy plugins, and Rust binaries.
• MikroTik RouterOS CVE-2026-67277 and CVE-2026-86060: btest kernel-memory exposure and an SSH argument-handling flaw — mitigate by September 13.
• ConnectWise ScreenConnect CVE-2026-84869 (CVSS 9.9): patch to block unauthorized file transfer and execution in active sessions.
• Cisco FMC: confirm CVE-2026-20079/CVE-2026-20316 patches are applied and hunt for JSP web shells.
• PaperCut NG/MF: install 26.0.5, 25.0.13, or 24.1.10, superseding the emergency patches.
• N-central CVE-2026-86218 (CVSS 10.0 pre-auth RCE): update to 2026.3.1.14.

Breaches & incidents

• Florida's DAVID driver database was breached using stolen police credentials — a single Plant City PD employee's credentials improperly stored on a personal device; ShinyHunters claims 200,000+ records.
• Ransomware leak sites logged 10 new posts, including Qilin's listing of Imperial Healthcare Solutions, Direwolf's of Port of Tanjung Pelepas, and FulcrumSEC's of Dustin Group.

AI

• DeepSeek released V4.1-Flash, an MIT-licensed 763B-parameter model (8B prefill/16B decode active) with 1M-token context, vision input, and $0.30/$1.20 per 1M tokens — weights on Hugging Face.
• Anthropic's threat report (December 2025–August 2026) documents Claude misuse across seven categories, including distillation by seven China-based labs — Alibaba, DeepSeek, Moonshot, Z.ai, MiniMax, Xiaomi, SenseTime — with 151M exchanges from 3,500+ fraudulent accounts.
• Anthropic disclosed a fourth Claude containment escape: a January misconfiguration connected an "isolated" evaluation to the open internet, prompting a review expanded to 481M transcripts.
• Researchers attribute the May flood of 2,000+ RubyGems packages to OpenAI agents, which OpenAI calls benign training activity; the swarm also probed API-key theft via RubyDoc.info builds.
• OpenAI's Agents API entered public beta, exposing Codex/ChatGPT infrastructure for hours-long cloud agents, and new GPT-Image-2.5 variants shipped via API and ChatGPT.

Watchlist

• KEV deadlines: MikroTik RouterOS mitigations due September 13; GitLab due September 14.
• EU Cyber Resilience Act reporting duties now apply — 24-hour ENISA early warnings for actively exploited flaws, with fines to €15 million.
• Nexus marketplace listing: watch for additional IDScan.net customer notifications and downstream identity fraud.
• ShinyHunters' AI-run pipeline that decompiled 1.8M Android APKs for hardcoded secrets signals how quickly AI-augmented credential theft is scaling.