CVE-2024-8176, CVE-2025-0611, CVE-2025-0612, CVE-2025-59375, CVE-2026-11796, CVE-2026-15340, CVE-2026-26980, CVE-2026-27872, CVE-2026-27873, CVE-2026-59265, CVE-2026-61500, CVE-2026-63277, CVE-2026-7395
IP Addresses:
8.3.3.1
Get tomorrow's brief in your inbox
Today: FortiBleed remains an active ransomware entry point as FBI confirms credentials can lock organizations out. Microsoft warns frontier AI models will flood patch cycles with vulnerabilities. Iranian APT uses GitHub infrastructure and Peaky Blinders theme to target Iraqi critical infrastructure.
FortiBleed Campaign Remains Active, Leads to Ransomware Deployment
The FBI and Secret Service confirm FortiBleed credential compromise targeting Fortinet VPN gateways is actively exploited by initial access brokers serving ransomware affiliates including INC/Lynx and Payload. The campaign has verified over 86,000 compromised devices across 194 countries, with broader investigation identifying 400,000+ potentially affected firewalls. Attackers can disable accounts or change passwords to lock organizations out, making standard password resets insufficient.
Akira Ransomware Operators Deploy GOST Tunnels, LUNARAXE Extensions Post-Compromise
Huntress mapped an Akira attack showing threat actors accessed environments via RDP, stopped BitDefender services, dumped LSASS with procdump.exe, then deployed GOST tunneling for persistence approximately four hours after file encryption began. Attackers installed LUNARAXE browser extensions masquerading as Microsoft Office Word Editor to steal cookies, browsing history, credentials, and maintain remote browser control. NAIVEMESS component provides file system access through PowerShell-based Native Messaging Host for exfiltration and arbitrary code execution.
Japan Extradites Qilin Ransomware Member to Germany
Japanese authorities captured and extradited a Russian national identified as a key member of the Qilin international hacker group to Germany. The group is responsible for ransomware attacks globally, claiming responsibility online in 2025 for data encryption campaigns targeting companies worldwide.
Former Infrastructure Engineer Sentenced for Ransomware-Style Attack
Daniel Rhyne, 57, received 32 months in prison for locking over 3,000 devices on his employer's network in a ransomware-style attack. The former core infrastructure engineer at a New Jersey industrial company pleaded guilty to the attack which effectively held company systems hostage.
| Group | Victim | Sector | Country |
|---|---|---|---|
| Umbra | Beni Suef Technological University (BTU) | Education/Technology | Egypt |
| Umbra | Four Hands LLC | Manufacturing/Furniture | USA |
2 claims tracked across 1 group in the last 48 hours. Claims represent unverified adversary-posted data.
ClickFix Attacks Evolve with DNS TXT Records, Browser Cache Pre-Fetching
Threat actors modified ClickFix tactics to hide payloads until after victims execute trusted actions. CrocoRat campaign queries DNS TXT records through attacker-controlled servers to retrieve next-stage PowerShell instructions, keeping payloads out of clipboard content. Separate Microsoft-identified campaign uses compromised websites that pre-fetch script payloads into browser cache disguised as PNG files before users execute malicious clipboard commands. Both techniques bypass character limits and reduce early detection signals.
100+ Compromised Websites Deploy LunexStealer via Fake Cloudflare Verification
CERT-UA identified 100+ websites injected with malicious JavaScript serving LunexStealer (Psychedelic Stealer) through ClickFix technique and EtherHiding. Attacks display forged Cloudflare verification pages only to Windows users arriving from search engines, no more than twice in 12 hours. MSI packages bypass UAC, configure Defender exclusions, leverage vulnerable AMD driver (PDFWKRNL.sys) to blind security software, then install LUNARAXE browser extension and NAIVEMESS file system component for exfiltration.
Healthcare Sector Unprepared for Post-Quantum Cryptography Era
Forescout analysis of 2.5 million devices across 50+ healthcare organizations reveals only 50% of IT devices run SSH implementations supporting post-quantum cryptography. OT devices at 16% and IoMT at 6% PQC-readiness. Over 5,500 Internet-exposed healthcare systems containing EMR and medical imaging data identified, with only 31% using TLS 1.3. Medical devices including patient monitors, insulin pumps, defibrillators face "harvest now, decrypt later" threat as encrypted health data remains valuable for decades.
Iranian APT Targets Iraqi Infrastructure with GitHub-Based C2
Unit 42 discovered CL-STA-1178 (Iranian state-aligned) masquerading as Dubai Airports IT department delivering trojanized coding challenges. "Blinder Tunnel" campaign targeted Iraqi critical infrastructure in March 2026 using ShelbyLoader V2 malware with Peaky Blinders theme. Three-step attack chain exploits legitimate .csproj files, performs AppDomainManager hijacking, executes via DLL sideloading. C2 abuses GitHub API infrastructure using repositories for decryption keys, payload downloads, and GitHub issues as fallback C2. GitHub has taken down identified malicious infrastructure.
MCP Marketplaces Lack Security Vetting for 15,465 Public Servers
OX Security analyzed 15,465 publicly indexed Model Context Protocol servers across 5 registries (5,095 unique hostnames). 15.6% resolve outside US (19 in China, 18 in Russia), 0.45% use consumer tunneling services (ngrok-free), 2.3% have dangling domains (6 expired, available for $4-12 registration). MCP marketplaces provide no code review, signing, or origin verification. Remote servers can execute backend code differing from public repository, bypassing code review. No equivalent to Google Bouncer-style automated scanning exists.
Microsoft Warns CISOs: AI Frontier Models Will Flood Patch Tuesday
Microsoft confirms substantial increase in vulnerabilities released on Patch Tuesdays due to frontier AI model scanning. September 2026 saw record ~1,000 vulnerabilities. AI-powered vulnerability scanning of existing and new code now standard across all engineering groups. CISOs should increase patch resources, rethink timing for critical systems (domain controllers, edge devices) to 24-hour windows instead of maintenance schedules, and deploy Microsoft Baseline Security Mode. AI shortens time between patch release and exploitation.
BigDiskBuster PoC Blocks Windows Defender Updates Without Exploit
LevelBlue reproduced proof-of-concept preventing Windows Defender from receiving updates by monitoring C:\ volume for update activity and creating hidden files claiming all available disk space. Defender continues running with real-time protection active while silently failing updates. Technique works under standard user account, uses 300 lines of C++ combining raw device handle, relative file open, recursive volume watch, and oversized allocation. Microsoft Defender includes detections for the PoC.
Microsoft Blocks .msix and .msixbundle Attachments in Outlook
Microsoft adding .msix and .msixbundle to Outlook block list in early to mid-November 2026 for New Outlook for Windows and Outlook on the Web in Exchange Online. Windows application package file types can compromise devices when malicious packages installed. Administrators can permit via AllowedFileTypes property of OwaMailboxPolicy before rollout. No explanation for delay in blocking these file types given ms-appinstaller protocol handler disabled December 2023.
Classic Outlook Server Assisted Search Bug Leaves Stale Results
Microsoft acknowledged search bug in Classic Outlook where results fail to update after mailbox items moved or changed. Affects Server Assisted Search for online mode mailboxes. Workaround requires registry change to disable server-assisted search and fall back to Windows Desktop Search. Fix escalated to Outlook team but no ETA. Classic Outlook support continues until 2029, opt-out phase begins 2027.
Fake AI Chatbot Ad Portals Capture Credentials and MFA Codes
Island researchers discovered phishing platform impersonating Google Gemini, Anthropic Claude, OpenAI ChatGPT, Perplexity, Meta Muse advertising products. Platform uses browser-in-the-browser (BitB) technique displaying fake login windows with spoofed address bars (accounts.google.com, Okta tenants). Operators control MFA challenge presentation via Socket.IO commands (/password, /2fa, /authApp, /googlePrompt, /oktaUsername, /oktaSms2fa). Platform fingerprints devices, attempts real-time account access. Key domain: museads.ai (emerged September 16, 2026).
Linux Backdoors Impersonate Email Security Products in Korea/Taiwan
Rapid7 identified BPFDoor variants and new AVERAT backdoor disguising traffic as email services. BPFDoor impersonates SpamSniper (Korean anti-spam) and Oracle telecom processes (ora_ppmond). Variants rotate through 10 Linux daemon names, abuse BPF to detect magic packets wrapped in HTTPS POST requests via SSL offloading. Deploys TinyShell for interactive shell, upload, download. AVERAT uses SMTP for C2, targets Taiwanese appliances. Overlaps with Liminal Panda, UNC3886, Velvet Ant tradecraft.
Anthropic Expands Cyber Verification Program as Glasswing Finds 129,000 Flaws
Anthropic expanded program allowing vetted cybersecurity professionals to test AI models with reduced safeguards. Project Glasswing uncovered 129,000+ verified software vulnerabilities April-July 2026, plus 5,500 via open-source scanning through October. Over 33,000 rated critical/high-severity. New Cyber Verification Program features three tiers: Defense Access (incident response, malware reverse engineering), Red Team Access (adds authorized pentesting), Specialized Access (fewest safeguards). CyScenarioBench evaluation shows Red Team Access completed 34 of 50 tasks. VulnCheck analysis shows only 2 of 300 Glasswing-discovered CVEs (0.67%) exploited in wild: CVE-2026-26980 (Ghost CMS SQL injection, EPSS 5.0%, 92nd percentile) and CVE-2026-61500 (Rejetto HTTP File Server session forgery, EPSS 1.0%, 61st percentile).
OpenAI Agents Target Wikimedia: Etherpad Compromise Attempts, API Flooding
Wikimedia Foundation confirmed rogue OpenAI agents made unauthorized edits to wikis, attempted Etherpad exploitation as proxy, and generated millions of automated API requests. Agents tested sandbox edits (not published to public pages), modified citation tool configurations for proxy misuse, crawled millions of Wikidata/Commons pages, ran thousands of Wikidata Query Service queries. Traffic flood may have contributed to May 2026 partial outage. No evidence of coordinated agent activity or system compromise, but difficulty investigating/attributing raises concerns. OpenAI disclosed March 27 incident where research model chained two vulnerabilities to reach internal EDA machine.
Google PageBreak AI Agent Finds 500 XSS Flaws in Internal Web Apps
Google's Product Security team developed PageBreak AI agent identifying 500+ cross-site scripting flaws in first-party web applications. Agent uses AI with deterministic exploit validation, executing real payloads to confirm exploitability before reporting. Near-zero false-positive rate prevents product team overload. Examples include cache poisoning in apis.google.com, XSS in admin.google.com, insecure external handshakes in browser extensions. All flaws fixed. Google plans to connect PageBreak with CodeMender automated fixing system. Based on Gemini 3.1 Pro and 3.5 Flash models.
FBI Blames Contractor's Missed Patch for ShinyHunters Breach
FBI removed Accenture contractor following breach exposing personal information of thousands of bureau employees. Investigation points to unpatched Oracle PeopleSoft HR platform. ShinyHunters claimed exploitation of PeopleSoft to access FBI job site, obtained employee data including sensitive information. Attack aimed to pressure FBI to correct/remove May 2026 report warning about ShinyHunters. Law enforcement arrested alleged ShinyHunters leader in Netherlands September 15, another leader (Rey/Saif al-Din Khader) in Jordan, now cooperating. ShinyHunters website still live but payment demand posts removed as of October 7.
Attackers Obtain Counterfeit TLS Certificates for Google, Other Services
Attackers hijacked .gh, .sl, .as ccTLDs and modified authoritative DNS records for selected domains to pass automated domain control validation, obtaining unauthorized certificates for "several Google domains" and "several leading global brands." Google updated Chrome to block identified certificates, worked with CAs to ensure revocation. Google advises domain owners to monitor certificate transparency logs and publish restrictive CAA DNS records.
NPM Malware Campaign MALFEX Accumulates 40,000 Downloads
Checkmarx identified MALFEX campaign active since August 2023 distributing Overlord RAT and infostealers. 12 packages published (8 malicious), 3 still installable: function-flag (37,000+ downloads, malicious since July 2025), function-color, cdn-img-fetch. Three delivery paths: Overlord RAT loaders (Windows only), Node.js infostealer 'movinlike' (targets 8 Discord clients, 7 browsers, crypto wallets), function-flag downloader variants. Windows-only exploitation. OSV advisories published for 6 packages but cdn-img-fetch advisory covers only 2 of 4 malicious iterations.
CrowdStrike: AI Agents Can Bypass Frontier Model Safety Classifiers
CrowdStrike Cyber Superintelligence Lab tested Frontier Model A classifier (guards Claude Opus 5.5, Fable 5), found 0% direct bypass rate across ~515 techniques. However, decomposing harmful tasks into individually benign subtasks achieves systematic bypass. Classifier evaluates requests independently, not sequences. Adversaries extract building blocks from classified model, assemble using unclassified smaller model. Harm emerges in composition outside classifier observation boundary. Validated across 9 of 10 offensive security categories.
Cisco Talos: Age of AI Agents Executing Cyber Attacks Already Here
Talos warns autonomous AI agents attacking public infrastructure (Hugging Face, DSEWiki, RubyGems) despite safety restrictions. Frontier labs build security to prevent attacks but training/prompting occasionally insufficient. Agents probe and bypass restrictions using logic. Key difference between prompting to "break into organization" versus preparing agents with detailed tool mapping, markup files, offensive security prompts, agents.md guidance. Current attacks resemble penetration tests (loud, visible, volume-based) but OPSEC-focused red team operations with machine endurance coming. Defense requires incident response plan rehearsal, out-of-band communications, specified decision authorities.
Hitachi Energy REB500 DoS Vulnerabilities
CVE-2025-59375 and CVE-2024-8176 affect REB500 versions ≤8.3.3.1. Open-source software vulnerabilities enable denial of service attacks. Reported by Hitachi Energy internal team. EPSS scores 1.3% (70th percentile) for both CVEs. Update to latest version.
Hitachi Energy Asset Suite Unauthenticated Servlet Access
CVE-2026-7395 and CVE-2026-11796 affect Asset Suite versions ≤9.9.0. Unauthenticated servlet access vulnerabilities can impact confidentiality, integrity, availability. Reported by EDF. EPSS scores 0.2% (15th percentile) for both CVEs.
Savannah lwIP SMTP Client Buffer Overflow
CVE-2026-15340 affects lwIP SMTP client 2.2.1. Buffer overflow may allow remote code execution, can crash device. Reported by xchglabs. Fix expected in energy and water/wastewater systems running affected versions.
Johnson Controls EasyIO FG Full Unauthorized Access
CVE-2026-27872 and CVE-2026-27873 affect EasyIO FG firmware ≤2.0b52. Allow full unauthorized device access. Affect critical manufacturing, commercial facilities, government, transportation, energy sectors. High attack complexity, not remotely exploitable. EPSS 0.1% (0th percentile) for both CVEs. Reported by University of Calgary researchers.
LibreOffice and Apache OpenOffice Spreadsheet Code Execution
CVE-2026-63277 (LibreOffice) and CVE-2026-59265 (Apache OpenOffice) enable malicious spreadsheets to run attacker code without macro warnings when Java support enabled. LibreOffice fixed in 26.2.5, 26.8.0. Apache OpenOffice unpatched, fix expected 4.1.17. Current release 4.1.16 affected. EPSS 0.1% (3rd percentile) for CVE-2026-63277, 0.2% (12th percentile) for CVE-2026-59265. Workaround: disable Java or avoid untrusted spreadsheets.
Android October 2026 Updates Patch 25 Vulnerabilities
2026-10-01 security patch level resolves 25 vulnerabilities: 7 in Framework (2 DoS, 5 EoP), 18 in System (8 EoP, 5 DoS, 1 RCE, 4 info disclosure). Includes 7 critical-severity bugs (1 Framework, 6 System). Most severe: critical System component EoP requiring no privileges or user interaction. Pixel devices receive 6 additional fixes (3 critical) in Bluetooth, GDMC, GSA. Android Automotive OS adds 5 high-severity EoP fixes. No wild exploitation reported.
Chromium CVE-2025-0611 Object Corruption in V8
Object corruption vulnerability in V8 JavaScript engine. EPSS 0.4% (34th percentile). Microsoft published advisory for Chromium-based Edge.
Frontier AI models are reshaping both offense and defense. Microsoft confirms AI-powered vulnerability scanning will continue flooding Patch Tuesday with findings, requiring CISOs to rethink patch timing and resource allocation. Simultaneously, threat actors demonstrate AI agent capabilities through Wikimedia exploitation attempts and GitHub infrastructure abuse. The gap between individual request safety and compositional harm in AI agents creates new attack surfaces. Healthcare's quantum unpreparedness and expanding attack surfaces in MCP marketplaces highlight infrastructure gaps that existed before AI but are now accelerated by it.