💥 BREACHES & INCIDENTS
Warp discloses breach exposing AI agent logs and user telemetry data
Warp disclosed a breach via a third-party vendor that exposed agent decision logs and telemetry data of roughly 1% of its users.- Warp users of the agentic development environment (800,000+ users) — roughly 1% may have been impacted
- Exposed data includes agent decision logs and telemetry; unclear whether source code, prompts, or credentials were affected
- Attack exploited a third-party vendor in Warp's infrastructure, detected within hours and remediated; analysis suggests an 'agent swarm' using a sophisticated penetration and escalation technique
- CrowdStrike engaged for investigation, law enforcement notified, affected users contacted
Ukrainian Hive and Nefilim ransomware affiliate Artem Stryzhak sentenced to 70 months in US prison
Ukrainian national Artem Stryzhak was sentenced to 70 months in prison for his role in Hive and Nefilim ransomware attacks.- Artem Stryzhak, 36, a Ukrainian national, was sentenced to 70 months in prison for conspiracy to commit computer fraud tied to ransomware attacks from mid-2018 to late 2021.
- Victims included hospitals, government entities, law firms, a dentist office, an oil and gas transportation company, and other businesses; the Hive conspiracy attacked roughly 1,500 victims worldwide and caused over $100 million in losses.
- Stryzhak joined Nefilim as an affiliate in June 2021, paying administrators including Volodymyr Tymoshchuk 20% of ransom proceeds for access to the ransomware.
- Nefilim crew customized ransomware per victim with unique decryption keys and ransom notes, targeting companies in the US, Canada, and Australia with over $100 million in annual revenue and threatening to publish stolen data.
- Stryzhak was arrested in Spain in June 2024, extradited to the US in April 2025, and pleaded guilty in December 2025; co-conspirator Volodymyr Tymoshchuk remains at large with an $11 million reward.
🕵️ RESEARCH & DEEP DIVES
CyberStrikeAI with 44 AI agents used in attacks on Japanese organizations, TrendAI finds
TrendAI researchers have disclosed a CyberStrikeAI-driven attack that used 44 AI agents to automate the full attack lifecycle against at least 19 organizations, 16 of them in Japan.- Targets include organizations in Japan, Taiwan and Hong Kong; at least 19 identifiable organizations, including at least 16 in Japan, were subjected to validation activity in a short October 2026 window.
- CyberStrikeAI is an open-source, Go-based AI-native offensive security platform by GitHub user Ed1s0nZ integrating 100+ security tools; it was previously tied to a campaign compromising over 600 Fortinet FortiGate devices across 55 countries.
- The attack delegated reconnaissance, vulnerability discovery, exploitation, C2, persistence, data exfiltration and automated reporting across 44 AI agents, generating Japanese-language forms and fake login pages.
- Two related campaigns from the same TrendAI analysis used AI for LLM-assisted planning targeting AI workflow platforms and inference services, and for tool development in attacks against Cisco, Citrix, F5 and Yamaha edge devices.
- The CyberStrikeAI server at IP 212.11.64.250 showed the tool's banner on port 8080; Team Cymru tracked 21 unique IPs running it between January 20 and February 26, 2026, mostly in China, Singapore and Hong Kong.
Aokigahara infostealer adds dedicated module targeting AI assistant credentials and sessions
The Aokigahara (Aoki) infostealer has begun stealing authentication and session files from AI coding assistants like Claude, Codex, Cursor, and GitHub Copilot.- Targets Windows machines, stealing AI-related authentication files, session data, and configs from Claude, ChatGPT Desktop, Codex, Cursor, Windsurf, GitHub Copilot, Gemini CLI, JetBrains AI Assistant, Codeium, and Kiro.
- Also steals Git credentials, SSH private keys, and AWS, Azure, and Google Cloud credential/config files, plus browser passwords, cookies, crypto wallets, password managers, VPN/FTP clients, and messaging app tokens.
- Newer samples target .claude.json, Claude desktop cookies, ChatGPT cookies, Codex config.toml and .env files; older samples did not target AI files, indicating a recent capability addition.
- Malware has no backdoor command list, collects whatever data is available, checks for VM/sandbox environments, and exfiltrates zipped, encrypted data to a C&C server via a /gate endpoint; older samples used unencrypted HTTP on port 8080.
- Symantec detects it as Infostealer.Aoki; sample IOCs include 752c8e64af60960b85cf00e4bd34eb459e24b7939464f718ec18494df43b4b69 and bf037920d39b85572ec7e53b51f09c28198c2c0afbc7c7d6cf639b1f047caebf.
📋 ADVISORIES
Microsoft releases Windows OOB updates to distrust compromised eMudhra CA certificates
Microsoft has released out-of-band Windows updates removing trust in certificates tied to compromised eMudhra CA keys.- Affects Windows 11 24H2/25H2/26H2 (KB5134133) and Windows 10 21H2/22H2 (KB5134125), especially systems without internet or with disabled CTL updates
- Eight eMudhra intermediate CA private keys generated in April 2025 were so weak (values under 256) they could be recovered almost instantly, allowing forged TLS and code-signing certificates
- Six compromised intermediates were used for TLS and two for code signing, chaining to roots trusted by Mozilla, Microsoft, Apple, and Chrome, including Sectigo's AAA Certificate Services
- The October 26 OOB update removes seven certificates from the Windows trust list: six eMudhra G3 roots and an AAA Certificate Services cross-sign; Edge and WebView2 also received CRLSet blocks
- Non-browser clients (Python, Ruby, Go, Node.js runtimes and bundled trust stores) largely don't check revocation, so only trust store updates remove the risk
📄 Source for The Gentlemen RaaS: rapid-scaling Qilin splinter group with 90% affiliate split — go.halcyon.ai
📄 Source for Critical Brotli decoder flaw (CVE-2026-111497) enables RCE across Node.js, Python, Bun, Envoy — github.com
🔓 CVEs & KEV
- CVE-2026-88812 — X.org xorg-x11-server — CVSS 7.8 — X.Org Server XKB SetGeometry TextDoodad Double FreeA double free in XKB SetGe...