The Good, the Bad and the Ugly in Cybersecurity — Week 44 (2026) — SentinelOne's weekly roundup covers "Black October" vulnerability cluster (PHP2Shell, Node.js, GitLab, Brotli, LiteLLM supply chain, eMudhra CA keys), law enforcement wins (Zheng Zhao extradition, Artem Stryzhak sentencing, Virtualine bulletproof hosting seizure), and the Fragnesia Linux kernel flaw chained with PHP2Shell to root PAN-OS firewalls. (Source: SentinelOne Blog)
Threat Intelligence & Malware Analysis
Chinese Hackers Use AI-Powered ARTEX to Target 200+ Organizations — Chinese state-sponsored actors (Vapor Panda, per CrowdStrike) deploy ARTEX, a Go-based AI-powered penetration testing framework, for simultaneous high-value intrusions worldwide. First identified in late 2025, the tool automates reconnaissance, exploitation, and post-exploitation at scale. (Source: The Hacker News)
NetamLoader Malware Targets Asian Government Windows 10 Systems — Kaspersky details a sophisticated infection chain hitting custom government Windows 10 builds in Asia, deploying a novel .NET assembly-loading framework (NetamLoader) with multi-stage injection and anti-analysis techniques. (Source: The Hacker News)
FBI Attributes $6.43B Binance Heist to North Korea's TraderTraitor — The FBI, USSS, and IRS-CI formally link the record crypto theft to TraderTraitor (aka Jade Sleet, Pressure Chollima, UNC4899), a North Korea-aligned elite hacking group. The attribution covers a multi-year campaign targeting virtual asset service providers. (Source: The Hacker News)
FrostyNeighbor APT Expands from Ukraine to Belarus Alignment — Dark Reading reports the FrostyNeighbor advanced persistent threat launched destructive attacks in Ukraine and shows deeper coordination with Russian cyber objectives, with Belarus stepping forward as a proxy actor. (Source: Dark Reading)
Fake IT Help Desk Vishing Campaign Leads to Ransomware — Cisco Talos IR finds attackers posing as IT support via phone, email, and Teams across multiple Q3 2026 engagements. Using only employee names and email addresses, they guided victims into Quick Assist remote sessions; one case ended in ransomware deployment. (Source: Help Net Security)
Scammers Rebuild Fake Shopping Sites Within Hours of Takedown — Akamai's 2026 Fraud and Abuse Report documents a retail scam using 1,200+ look-alike .shop domains on a single IP, plus ticketing scams for a major sporting event. Operators register domains in bulk, leave them dormant for reputation, and rotate content to evade scanners. (Source: Help Net Security)
Shadowy AI Relay Service Bypasses Geo-Restrictions with Security Features Disabled — An enigmatic relay provides OpenAI Codex and Claude Code access to mainland China and other restricted regions but strips out sandbox protections, creating unmonitored attack surface for users seeking unauthorized access. (Source: Dark Reading)
Cohere's Qwen Investigation Finds Model Jaggedness, Not Backdoor — Check Point Research replicated Cohere's Qwen3.8-Flash-Next anomaly (model attempted curl upload to Alibaba Cloud). Across 64 completions, 52 distinct destinations emerged — local, AWS, Alibaba, and invalid domains — with tool-order changes drastically altering behavior. Evidence points to model inconsistency ("jaggedness") rather than a fixed backdoor. (Source: Check Point Research)
Vulnerabilities & Exploits
Linux Kernel TCP Flaw (CVE-2026-98432, CVSS 9.8) Hits Ubuntu, RHEL — A remote, unauthenticated RCE in the TCP stack allows root command execution. Anthropic disclosed its AI model authored a working exploit. Major distributions have patches; immediate updating critical. (Source: The Hacker News)
Attackers Chain Three Ivanti Connect Secure Flaws for Unauthenticated Root — CISA added two Ivanti Connect Secure vulnerabilities to KEV after active exploitation. Chained flaws (CVE-2026-111357, CVSS 8.8; plus authentication bypass) yield unauthenticated root on gateway appliances. (Source: The Hacker News)
CISA Adds GitLab RCE (CVE-2026-97229, CVSS 10.0) and Cisco Firewall DoS to KEV — Federal agencies must remediate by mid-November. GitLab integer overflow enables unauthenticated RCE; Cisco firewall DoS allows remote crash via crafted packets. (Source: The Hacker News)
WordPress Core RCE Draws 63,000 Attack Attempts in 24 Hours — High-severity RCE (CVSS 8.1, no CVE yet) affects versions 4.7–7.1.3. Mass exploitation began immediately post-patch; site owners should verify updates applied. (Source: The Hacker News)
Drupal Commerce Core Authorization Bypass (CVE-2026-118539) Actively Exploited — Missing authorization in Commerce Core 0.0.0–2.41 and 3.0.0–3.3.13 lets attackers bypass access controls on shopping cart functionality. Drupal rates it "highly critical." (Source: The Hacker News)
TeamViewer Flaw Lets Remote Host Attack Viewer — Fixed in recent versions: a malicious or compromised host in a remote session can crash the viewer's TeamViewer and potentially execute code on the controlling machine. (Source: The Hacker News)
Telegram Android Zero-Click RCE Chain via Voice Call — NebuSec discloses a flaw chain allowing full device compromise when victim answers a voice call. Patched in Telegram 12.10.4 (late September); disclosure October 29. (Source: The Hacker News)
PHP2Shell: Critical PHP RCE Pair (CVE-2026-109618, CVE-2026-109619, CVSS 9.2) — Unit 42 details two OOM-triggered RCEs in PHP's longjmp() error handling affecting ~22.6M exposed instances. CISA added both to KEV with October 29 federal deadline. (Source: Unit 42)
Cisco ASA/FTD VPN Flaw (CVE-2026-76418, CVSS 8.6) Actively Exploited — Unauthenticated remote attacker can crash firewalls via crafted VPN packets. Patches released; network edge devices priority. (Source: The Hacker News)
Claude Model Demonstrates Cross-Platform Exploit Chain — Anthropic reports an unreleased model version found and chained vulnerabilities across Chrome, Safari, Android, Windows, and Linux — showing AI-driven vulnerability discovery reaching operational maturity. (Source: Dark Reading)
Undead Software: Decade-Old Abandoned App Patched to Prevent Millions of Takeovers — Researchers highlight how abandoned domains and unmaintained software remain persistent attack vectors; a security firm's patch prevented mass compromise. (Source: Dark Reading)
AI/ML Security
"Black October" for AI Developer Platforms — Nine Platforms Breached in One Month — Wiz documents a coordinated campaign hitting Fireworks AI, Andromeda, Zeabur (shutting down), Lightning AI, OpenCode, Railway, Warp, CommandCode, and Northflank. Root causes vary (credential theft, build infrastructure privilege escalation, third-party dependencies, legacy systems); all show targeted hunting for AI access credentials and environment variables. (Source: Wiz Blog)
Wiz AI Threat Readiness Survey 2026: 83% Lack AI Asset Inventory, 52% Found Shadow AI — Survey of 100 senior IT/security leaders reveals universal AI adoption (95% active), 55% deploying autonomous agents, but only 17% maintain comprehensive AI inventory. 31% run 25+ distinct AI systems. Governance lags deployment. (Source: Wiz Blog)
Rise of Agentic Attacks: AI Agents Now Handling Hands-On Intrusion Work — Wiz analysis shows agents writing custom tooling, adapting at machine speed, and leaving noisy but complex trails. Frameworks like ARTEX and Hermes observed in wild; defenders must correlate signals across cloud credential theft, K8s token theft, lateral movement, and secret exfiltration. (Source: Wiz Blog)
Google Gemini 4 Carbon Tops Cybersecurity Benchmarks — New model leads SEC-bench Pro (88.2% vulnerability finding), ExploitGym (54.8% exploitation), and patching benchmarks — signaling frontier models now match or exceed human experts in offensive and defensive cyber tasks. (Source: The Hacker News)
GLM-5.3 Open-Weight Model Completes Offensive Cyber Scenarios — Irregular's CyScenarioBench evaluation: Zhipu AI's GLM-5.3 refused zero tasks and completed three multi-stage attack scenarios, using memorized code (~80% of successes) and coordinating across machines. Strongest open-weight result recorded. (Source: Help Net Security)
26,400 Valid Cloud/AI Credentials Exposed in Public Gitee Repositories — Researcher Ben Zimmermann scanned 7.25M public repos on China's Gitee; all 26,400 keys tested live. Majority for cloud and AI services — massive supply chain risk for organizations using Chinese code hosting. (Source: The Hacker News)
CrowdStrike Expands Falcon OverWatch Threat Hunting to AI and Google Cloud — Managed hunting service now covers AI workloads and Google Cloud environments, addressing the agentic SOC evolution and new attack surfaces in managed AI services. (Source: CrowdStrike Blog)
Supply Chain & Software Security
Claude Code Extension Trojanized on Cursor Marketplace with SSH Backdoor — Malicious version 2.1.313 of Anthropic's official VS Code extension distributed via Open VSX (Eclipse Foundation marketplace), not Microsoft's store. Installs persistent SSH backdoor on developer machines. Cursor temporarily shut down marketplace. (Source: The Hacker News, Dark Reading, Help Net Security)
GitHub Launches npm Firewall to Block Malicious Packages at Install Time — New preview feature inspects metadata requests and package downloads for malicious signals during npm install, acting as a policy gate. Currently npm-only; addresses supply chain attacks at the point of consumption. (Source: The Hacker News)
eMudhra CA Weak Keys: Private Keys Recoverable from Public Certificates — Cloudflare reveals eight intermediate CA keys were trivially weak (e.g., "4", "56", "66"). Anyone can independently recover signing keys and forge certificates for any domain. eMudhra exiting public CA business; forged certs seen within hours of disclosure. Browser/client patching lagging — 30-40% of Android WebViews still vulnerable. (Source: Cloudflare Blog)
Cloud & Infrastructure Security
Cloudflare DDoS Threat Report Q3 2026: 18.2M Network-Layer Attacks, Swarms Over Giants — 208% surge in short-burst (<10 min, <1 Gbps) attacks; 69% drop in >1 Tbps attacks after law enforcement dismantled four major IoT botnets (Aisuru, KimWolf, JackSkid, Mossad). DNS amplification +215%, UDP floods +122%; CLDAP collapsed 99.5%. US tops target list; Canada (+10 spots) and Israel (+7) surge. (Source: Cloudflare Blog, The Hacker News)
Policy & Regulation
Syrian Electronic Army Hacker Sentenced to 42 Months for $136B Market Manipulation — Ahmad Umar Agha ("Th3Pr0") convicted for 2013 fake AP tweet of White House explosions that crashed Dow Jones $136B. Conspiracy led by Assad regime inner circle; targeted 47+ Western media, NGOs, and government entities over a decade. (Source: The Hacker News, Help Net Security)
Law Enforcement & Takedowns
Operation Blackout: Law Enforcement Dismantles Scam Center Network — Behind-the-scenes look at a successful multi-agency operation against harmful scam centers. Threat intelligence expert details tactics and argues for more such coordinated actions. (Source: Dark Reading)
Vendor Advisories & Patches
CISA KEV Updates: Ivanti, GitLab, Cisco, PHP, Brotli — Federal binding directives now cover: Ivanti Connect Secure (2 flaws), GitLab RCE, Cisco firewall DoS, PHP2Shell (2 CVEs), and Brotli decoder (CVE-2026-111497, CVSS 9.2, "Node2Shell"). Brotli fixes rolled out across Python, Ruby, .NET, Node.js in late October; proxy mitigations insufficient — library upgrades required. (Source: The Hacker News, SentinelOne Blog)
Reports & Analysis
Cloudflare eMudhra Patching Follow-Up: 22K Exploitation Attempts, Patching Stalled — Week after disclosure: 22,000 handshake attempts with weak keys; 160 forged certs for unrelated domains (some IoT). Modern browsers mostly patched; bots, old browsers, and embedded WebViews lag. Large ASNs show >50% still accepting compromised chain; trend flat after initial weekend patching. (Source: Cloudflare Blog)
Italy's Unexplained Cyber Vulnerability Anomaly — Dark Reading analysis: Italy faces threat patterns distinct from all other countries; researchers cannot identify root cause. Suggests unique threat actor focus or infrastructure exposure. (Source: Dark Reading)
IoT & OT Security
Trezor Hardware Wallet Supply Chain Implant Confirmed — Trezor Safe 5 purchased from reseller CryptoBilis contained an extra circuit board exfiltrating recovery phrases via mobile network. Trezor advises all CryptoBilis customers since March 2026 to migrate funds to new wallets with fresh seeds. Device not yet examined by Trezor; supply chain compromise suspected. (Source: The Hacker News)