PHP2Shell, ARTEX AI Campaigns, and Edge Appliance Zero-Days Define October 2026 Threat Reporting
October 2026 threat reporting centered on the disclosure and rapid exploitation of two critical unauthenticated remote code execution flaws in PHP core (CVE-2026-109618 and CVE-2026-109619), multiple state-aligned and criminal actors weaponizing the ARTEX agentic AI penetration testing framework to automate intrusions against hundreds of organizations, and active exploitation of zero-day vulnerabilities in Ivanti Connect Secure, BeyondTrust Remote Support, and Atlassian Data Center products. A Brotli decoder vulnerability (CVE-2026-111497) affected Node.js and numerous downstream libraries, while GitLab disclosed a CVSS 10.0 unauthenticated RCE in its CI/CD regex compiler. Supply chain compromises extended across npm packages, GitHub Actions workflows, Terraform providers, and VS Code extensions. Ransomware operators including n0n, INC, and Aur0ra incorporated AI tools into their attack workflows. State-sponsored espionage campaigns from APT28, Lazarus, APT41, and Kazakhstan-nexus UAT-6275 targeted defense, diplomatic, and government organizations. Device code phishing kits multiplied against Microsoft 365 tenants, and infostealers began harvesting credentials from AI coding assistants including Claude, Cursor, and GitHub Copilot.
Improve this report
Send corrections, useful missing information, or quality feedback. It is private and will not appear as a comment.
Detection / HunterZ.ai
Themes and developments
PHP2Shell: Unauthenticated RCE in PHP Core
Wiz Research discovered two use-after-free and memory corruption vulnerabilities in PHP's session.upload_progress multipart handling that enable unauthenticated remote code execution on any PHP server serving web requests. CVE-2026-109618 corrupts memory during out-of-memory handling via longjmp(), and CVE-2026-109619 reuses stale internal request state across worker requests. Both flaws carry CVSS v4.0 scores of 9.2 and affect PHP versions from 5.4 through 8.5, with patches released October 21, 2026 for branches 8.2 through 8.5 (Kaspersky analysis, Talos analysis).
Approximately 5.1 million PHP IP addresses were exposed per VirusTotal scans, and roughly half of PHP deployments run end-of-life branches with no available patch (Recorded Future). A public proof-of-concept exploit named PHP2Shell circulated within hours of disclosure, with 99 GitHub repositories appearing the same day, 83% of community interest directed at offensive PoCs rather than defensive tools (AhnLab). Amazon honeypots captured exploitation on October 17, four days before the fix was released (Sekoia).
CISA added both CVEs to its Known Exploited Vulnerabilities catalog on October 26 (CISA). Arctic Wolf linked active scanning and weaponized payload delivery to the Potassium botnet, with payloads hosted at 92.118.112.100 (Arctic Wolf). Morphisec documented the first PHP2Shell exploitation against Windows IIS servers running PHP-CGI, deploying VULCANRAT via PEB-walking shellcode (Morphisec). CrowdStrike assessed that per-deployment calibration of heap offsets and memory grooming would constrain near-term mass exploitation despite widespread attention (CrowdStrike).
Sources: 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19, 20, 21, 22
Reporting chronology
| Publication date | Reported update | Source |
|---|---|---|
| 2026-10-21 | PHP Foundation released patches for two CVSS 9.2 unauthenticated RCE flaws affecting all versions from 5.4 onward | Source |
| 2026-10-21 | Kaspersky and Talos published technical analysis confirming mod_php highest risk and PoC circulation | Source |
| 2026-10-22 | Arctic Wolf linked active scanning and weaponized payloads to Potassium botnet infrastructure | Source |
| 2026-10-26 | CISA added both CVEs to Known Exploited Vulnerabilities catalog confirming active exploitation | Source |
| 2026-10-28 | Morphisec documented first PHP2Shell exploitation against Windows IIS servers running PHP-CGI | Source |
ARTEX AI-Driven Intrusion Frameworks Scale Offensive Operations
CrowdStrike identified an unknown Chinese-speaking financially motivated threat actor using ARTEX, an open-source Chinese agentic AI penetration testing framework, to exploit South Korean financial institutions between September and October 2026. The framework used DeepSeek, GLM-5.3, and Grok 4.6 as LLM backends, exfiltrating PII of approximately 67,000 individuals from loan inquiry and employee mobile work systems (CrowdStrike). AhnLab identified 600+ internet-exposed ARTEX instances, some containing web shells and pass-the-hash tools, and confirmed in controlled tests that ARTEX autonomously exploited CVE-2023-46604 in Apache ActiveMQ and achieved lateral movement to four internal hosts without human intervention (AhnLab).
Recorded Future attributed a broader ARTEX campaign to TAG-217, which operated from August to October 2026 targeting at least seven South Korean financial institutions and exposing data from at least 68,000 individuals. TAG-217 also tested iPhone and Android exploitation techniques and targeted YiiMP cryptocurrency mining platforms (Recorded Future). Microsoft linked Storm-0940, a China-state aligned actor active since 2021, to ARTEX-driven password spraying and OAuth application abuse against ConnectWise multi-tenant credentials, enabling app-only authentication into over 2,250 customer Microsoft 365 tenants with privileged access achieved in under three hours in multiple cases (Microsoft.
CrowdStrike and Mandiant reported that VAPOR PANDA (aka Earth Estries/Salt Typhoon) used ARTEX powered by DeepSeek to conduct simultaneous intrusions against 200+ organizations across 14 countries starting October 17, 2026, deploying the BLOODALCHEMY backdoor via DLL sideloading and consuming 1.4 billion input tokens in a single intrusion (CrowdStrike, Mandiant). Mandiant also observed UNC7228, a PRC-nexus ARTEX operator overlapping with Storm-0940, achieving deep system access in approximately 15 minutes in one incident. Multiple ARTEX forks including ScopeWeaver, Tangora PTaaS, and ARTEX PRO were identified after the original creator took the repository closed source in October 2026.
Sources: 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14
Reporting chronology
| Publication date | Reported update | Source |
|---|---|---|
| 2026-10-14 | AhnLab found 600+ exposed ARTEX instances and confirmed autonomous lateral movement in lab tests | Source |
| 2026-10-16 | Recorded Future attributed ARTEX campaign to TAG-217 hitting 68,000 individuals across seven financial institutions | Source |
| 2026-10-21 | Microsoft linked Storm-0940 to ARTEX-driven ConnectWise credential compromise affecting 2,250 tenants | Source |
Ivanti Connect Secure Zero-Day Exploitation Chain
Volexity disclosed a chained exploitation of previously unknown flaws in Ivanti Connect Secure VPN appliances on October 13, 2026. The chain combines an authentication bypass in welcome.cgi, a race-condition RCE in wu.cgi, and a privilege escalation via dsinvoked DSCache manipulation, enabling unauthenticated remote command execution without deploying a persistent backdoor. Post-exploitation targets included LAPS passwords, BitLocker keys, and LDAP credentials, with anti-forensics patching libraries in memory only, evading the Ivanti Integrity Checker (Volexity). CISA confirmed active exploitation and added CVE-2026-104935 to the KEV catalog the same day, noting that the Integrity Checker Tool may not detect in-memory payloads (CISA).
Volexity assessed with medium confidence that actor UTA0576 used LLMs for exploit development and post-exploitation. A separate actor, UTA0533, was noted using similar LDAP packet capture techniques on SonicWall SMA VPN appliances. The patch for CVE-2026-104935 was released October 13, but a privilege escalation fix remained pending at the time of disclosure. On October 30, CISA added two additional Ivanti Connect Secure CVEs to the KEV catalog: CVE-2026-111357 (missing authentication for critical function) and CVE-2026-118527 (predictable identifiers or tokens), both under active exploitation (CISA).
Reporting chronology
| Publication date | Reported update | Source |
|---|---|---|
| 2026-10-13 | Volexity disclosed zero-day chain in Ivanti Connect Secure with active exploitation by UTA0576 | Source |
| 2026-10-13 | CISA added CVE-2026-104935 to KEV catalog confirming active exploitation for credential theft | Source |
| 2026-10-30 | CISA added CVE-2026-111357 and CVE-2026-118527 revealing additional actively exploited Ivanti flaws | Source |
Atlassian Pre-Auth File Read and BeyondTrust Authentication Bypass Exploited
WatchTowr disclosed CVE-2026-21589, a pre-auth arbitrary file read across eight Atlassian Data Center products via the unescapeSlashes function in the shared atlassian-plugins-webresource library. The flaw uses double-colon encoding to bypass slash filtering, and when Crowd SSO is configured, attackers can read plaintext application credentials from crowd.properties and create administrator accounts through Crowd's REST API. Approximately 700,000 Confluence instances are exposed to the internet (WatchTowr). CISA confirmed active exploitation and the CVE was added to the KEV catalog. Atlassian Cloud was already patched with no evidence of exploitation (CERT-EU, Akamai).
Mandiant reported that ARMORY CASTLE, a suspected PRC-nexus espionage actor previously tracked as UNC5221, exploited CVE-2026-102764 in BeyondTrust Remote Support (formerly Bomgar) to deploy the SAWFLY PHP webshell and BARECLAW SOCKS5 proxy. The flaw allows unauthenticated file operations on self-hosted appliances up to version 26.2.3, with persistence achieved via sudoers template modification surviving reboots. SaaS deployments were patched October 9, 2026 (Mandiant). CISA added CVE-2026-102764 to the KEV catalog on October 16, noting active exploitation involving initial access, persistence, privilege escalation, and lateral movement (CISA).
Sources: 1, 2, 3, 4, 5, 6, 7, 8
Reporting chronology
| Publication date | Reported update | Source |
|---|---|---|
| 2026-10-06 | WatchTowr published technical analysis of CVE-2026-21589 pre-auth file read across eight Atlassian products | Source |
| 2026-10-15 | Mandiant attributed ARMORY CASTLE to BeyondTrust exploitation with SAWFLY webshell and BARECLAW proxy | Source |
| 2026-10-16 | CISA added CVE-2026-102764 to KEV confirming active BeyondTrust exploitation with lateral movement | Source |
ConnectWise Cloud Backup Cross-Tenant Credential Abuse
Huntress reported that stolen ConnectWise Cloud Backup (formerly SkyKick) service principal credentials were used from IP address 144.172.64.128 on October 15, 2026 to authenticate across customer Microsoft 365 tenants using the app's own identity. An automated script enumerated consenting tenants in ascending GUID order over 96 minutes, accessing 162+ tenants. In one tenant, a related ConnectWise app assigned itself Global Administrator and added persistent client secrets expiring 2028 to other applications, including a security platform (Huntress).
Sophos confirmed the multi-tenant Entra application's service principal authenticated across customer tenants using app-only authentication, with no evidence of token misuse or post-access activity in reviewed tenants. The sweep included tenants with no active backup subscription (Sophos). Microsoft attributed the ConnectWise credential compromise to Storm-0940 using ARTEX to automate password spraying and OAuth application abuse at machine speed, with time from initial targeting to privileged access under three hours in multiple cases (Microsoft. By October 19, Check Point reported 173 organizations affected (Check Point.
Supply Chain Compromises Across Developer Ecosystems
Socket reported that [email protected], an npm package receiving approximately 12,000 weekly downloads, was published with a preinstall hook executing an obfuscated loader that harvested tokens from npm, GitHub, AWS, Kubernetes, and HashiCorp Vault, republishing compromised packages using stolen credentials. The worm resolved C2 via an Ethereum contract using approximately 30 RPC endpoints, matching the August ChainDrop pattern seen in keyv and cacheable packages (Socket). A separate npm campaign using indexed-btree packages shifted to import-time execution, firing malicious code when applications called the library rather than at install time, defeating npm v12's default of not running install scripts (Reversing Labs).
GhostAction compromised two GitHub maintainer accounts to inject malicious GitHub Actions workflows into 346 repositories, including uber/athenadriver and kitao/pyxel (18,420 stars). Unlike earlier waves, this variant used fetch-depth: 0 and git log -p --all to scan complete git history for AWS keys, AI service tokens, and source control credentials, POSTing harvested secrets in cleartext (Socket). GlassWorm-linked malicious VS Code theme extensions delivered staged malware loaders across VS Code Marketplace and Open VSX, with one variant using Solana transaction memos as dead-drop resolvers for C2 infrastructure (Socket).
Reversing Labs identified malicious Terraform providers on the HashiCorp registry typosquatting a provider with 56 million downloads, sharing blockchain and Slack C2 infrastructure with the npm Graphalgo campaign (Reversing Labs). Mandiant reported UNC7017 compromising AOMEI's update infrastructure through 5-10 minute configuration tampering windows, redirecting update downloads to CROSSROAD and SHINYROAD backdoor payloads via DLL search-order hijacking of signed Radmin binaries (Mandiant. A malicious Claude Code extension (v2.1.313) distributed through Cursor's marketplace for approximately 11 hours on October 29 added attacker SSH keys and reverse SSH tunnels to developer workstations, reaching up to 5,000 users (Socket, Microsoft.
Sources: 1, 2, 3, 4, 5, 6, 7, 8, 9, 10
Reporting chronology
| Publication date | Reported update | Source |
|---|---|---|
| 2026-10-02 | GlassWorm-linked malicious VS Code extensions found using Solana transaction memos for C2 resolution | Source |
| 2026-10-08 | Tensorlake npm package compromised with preinstall worm harvesting cloud and AI credentials | Source |
| 2026-10-09 | GhostAction compromised two GitHub accounts injecting credential-stealing workflows into 346 repositories | Source |
| 2026-10-15 | Mandiant disclosed AOMEI update infrastructure tampering with 5-10 minute windows delivering backdoors | Source |
| 2026-10-30 | Malicious Claude Code extension distributed via Cursor marketplace adding SSH backdoors to developer machines | Source |
Node2Shell: Brotli Decoder Vulnerability Across Multiple Runtimes
Akamai disclosed CVE-2026-111497, an out-of-bounds write in Google's Brotli decoder (c/dec/decode.c) caused by malformed block switches in compressed streams. The vulnerability carries CVSS 9.2 and affects Brotli versions 1.0.3 through 1.2.0, with downstream impact on Node.js, Bun, .NET, Envoy Proxy, Istio, Python Brotli, Java brotli4j, and curl. Node.js released fixes on October 21 for versions 22.23.4, 24.22.1, and 26.12.1 (Akamai). Express 5 with body-parser 2 decompresses Brotli request bodies by default, exposing applications that never explicitly handle Brotli.
CISA added CVE-2026-111497 to the KEV catalog on October 27 (CISA). Zscaler ThreatLabz observed 20 source IPs conducting reconnaissance and exploitation against 12 customers within 24 hours of disclosure, with Imperva demonstrating RCE in Node.js 24.22.0 Linux arm64 containers (Zscaler). AhnLab named the vulnerability Node2Shell and identified three attack entry points: HTTP POST bodies, external server responses, and direct file processing (AhnLab. Sophos confirmed active exploitation in the wild, noting Microsoft and PHP also published product updates for this CVE (Sophos. Discontinued Node.js lines (25.x, 23.x, 20.x, and earlier) receive no patch and require migration.
GitLab Critical Unauthenticated RCE Disclosures
CERT-EU reported CVE-2026-97229, a CVSS 10.0 unauthenticated remote code execution in GitLab CE/EE CI/CD regex compiler via integer overflow from crafted regular expressions. Fixed versions 19.4.2, 19.3.4, and 19.2.8 were released before October 13 (CERT-EU). On October 22, CERT-EU disclosed CVE-2026-110264, a separate CVSS 10.0 integer overflow in GitLab's regex compiler affecting every supported branch from 7.0 to 19.5, with patched versions in 19.5.1 and backported fixes. GitLab published two detection rules for exploitation attempts (CERT-EU). CISA added CVE-2026-97229 to the KEV catalog on October 29 (CISA). The Malpedia weekly recap noted that Ruby Onigmo regex engine CVEs CVE-2026-89078 and CVE-2026-93577 also affect GitLab servers (Malpedia.
NetScaler Mass-Exploitation Framework and Additional CVEs
Following September's disclosure of CVE-2026-88771 and CVE-2026-88772, Malpedia published analysis of the NetScaler C2 toolkit, an automated mass-exploitation framework with Chinese-language code comments that discovers exposed NetScaler devices via FOFA, fingerprints versions, and exploits CVE-2026-88771 across 11 endpoints. The flaw is a log injection where attacker-controlled text from unauthenticated HTTP endpoints is written to ns.log and later executed as root by the admautoregd daemon on a 24-hour cycle, causing delayed payload execution (Malpedia). Zscaler published independent analysis confirming the exploitation mechanism via PPE failure record processing, with web shells named Nx_7508 deployed across six directories and a download server shared with StyleSmuggler infrastructure (Zscaler.
CISA added CVE-2026-88779, a memory overflow in SAML-configured NetScaler appliances causing denial of service, to the KEV catalog on October 4 (CISA). Sophos confirmed the flaw affects only customer-managed appliances configured as SAML Service Providers or Identity Providers, with Citrix-managed cloud services already patched (Sophos. A distinct SAML authentication issue introduced in Update 1 caused system instability and reboots not remediated by existing patches. The Canadian Cyber Centre issued advisories on October 9 for CVE-2026-107406 affecting additional NetScaler builds (Cyber Centre.
Sources: 1, 2, 3, 4, 5, 6, 7, 8
Reporting chronology
| Publication date | Reported update | Source |
|---|---|---|
| 2026-10-04 | CISA added CVE-2026-88779 SAML memory overflow to KEV confirming active exploitation | Source |
| 2026-10-07 | Malpedia published NetScaler C2 toolkit analysis revealing automated mass-exploitation with FOFA discovery | Source |
| 2026-10-09 | Canadian Cyber Centre issued advisory for CVE-2026-107406 affecting additional NetScaler builds | Source |
Ransomware Operations Adopt AI for Attack Automation
Malpedia reported that INC ransomware and The Gentlemen ransomware affiliates used AI-assisted operations with OpenCode and commercial LLMs via OpenRouter, logging 8,257 AI messages across 133 sessions with over $23,920 in API spending. Operators exploited FortiClient EMS, SimpleHelp, and cPanel vulnerabilities for initial access, deployed custom Phantom and Ghost C2 frameworks, and staged 46 INC encryptor builds for six victims targeting 62 organizations primarily in manufacturing and healthcare (Malpedia). Malpedia separately documented n0n's AI-assisted extortion operation, where an exposed server leaked 1,729 files including attack scripts and encryptor sources. n0n relied on credential stuffing FortiGate VPNs and exposed AWS keys rather than software vulnerabilities, listing 93 organizations across 14 countries since September 2026 (Malpedia.
Sophos reported that the GOLD OGRE group operates Aur0ra ransomware using AI agents to automate intrusion phases from credential access through deployment, with Cursor's Composer agentic model orchestrating AD discovery, Kerberoasting, NTLM relay, and AD CS abuse with minimal human involvement. AI guardrails were bypassed by framing intrusions as authorized red team testing (Sophos. CISA published a joint advisory on SafePay ransomware, which has affected 600+ victims across four continents since August 2024 using ChaCha20-RSA hybrid encryption and exploiting five CVEs in edge devices for initial access (CISA. ANY.RUN documented IronChain ransomware, which destroys RSA private keys at generation, making file recovery impossible even with payment (ANY.RUN.
Sources: 1, 2, 3, 4, 5, 6, 7, 8, 9
Reporting chronology
| Publication date | Reported update | Source |
|---|---|---|
| 2026-10-01 | Symantec reported Longlegs deploying Warlock ransomware via SharePoint ToolShell against water and telecom operators | Source |
| 2026-10-20 | CISA published SafePay advisory covering 600+ victims with five edge-device CVEs for initial access | Source |
| 2026-10-20 | Malpedia exposed INC and Gentlemen AI-assisted operation with 8,257 AI messages and 62 targeted organizations | Source |
| 2026-10-28 | Malpedia documented n0n AI-assisted extortion with exposed operator workspace leaking 1,729 files | Source |
| 2026-10-29 | Sophos reported GOLD OGRE using Cursor AI agents to automate Aur0ra ransomware intrusions | Source |
State-Sponsored Espionage Campaigns Across Multiple Regions
Symantec reported that APT28 (Forest Blizzard), a Russian GRU unit, used the Hookedge backdoor via Word macros to intrude on a naval shipbuilder in a NATO country from July through late September 2026. Hookedge communicates via webhook.site using a hidden 1-pixel Edge browser instance for C2, with attackers escalating to a dedicated IP and deploying ld.exe, an undocumented reflective loader (Symantec. Kaspersky documented Lazarus Group's Operation FlashHole targeting South Korean organizations since July 2025 by abusing mandatory CrossWeb EX financial security software, with a multi-stage in-memory chain using GitHub API for C2 and experimental ML-KEM-1024 post-quantum key exchange (Kaspersky.
Kaspersky reported APT41 deploying BlackX, a UEFI bootkit derived from BlackLotus, against diplomatic organizations in Central Asia. BlackX exploits CVE-2022-21894 (Baton Drop) on fully patched Windows machines because the patch requires manual certificate revocation activation, then deploys a cloud-storage-backed RAT using Google Drive, OneDrive, and Yandex.Disk for C2 (Kaspersky. Talos disclosed UAT-6275, a Kazakhstan-nexus APT targeting CIS government entities across nine countries since June 2025 using four infection chains delivering custom Chvost, Laplas, and PickleShell reverse shells, with exposed GitHub repositories and FortiWeb PoC code (Talos. Recorded Future provided parallel coverage under the TAG-193 designation, noting the group's UTC+5 working hours and targeting of diplomacy, water resources, and energy (Recorded Future.
Unit 42 uncovered PEERSTITCH, a memory-resident ARM64 FortiGate implant deployed after attackers exploited FortiManager CVE-2024-47575 to steal configs and crypto material for 500+ firewalls at a telecommunications company. PEERSTITCH uses ptrace-based injection into miglogd and iked daemons with covert inter-process memory sharing and Apache mod_md ACME abuse for C2 (Unit 42. Talos separately reported UAT-11473 exploiting CVE-2026-20127 (CVSS 10.0) in Cisco SD-WAN to deploy KeyDrift malware that steals AES-256 IPsec encryption keys from decrypted control sessions, with infrastructure overlapping Volt Typhoon's Silent Train network (Talos.
Sources: 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12
Reporting chronology
| Publication date | Reported update | Source |
|---|---|---|
| 2026-10-08 | NCSC and international partners called out Integrity Technology Group for AI-assisted scanning and botnet operations | Source |
| 2026-10-14 | Talos disclosed UAT-6275 Kazakhstan-nexus espionage targeting CIS governments with custom reverse shells | Source |
| 2026-10-20 | Symantec reported APT28 Hookedge backdoor intruding on NATO naval shipbuilder since July 2026 | Source |
| 2026-10-21 | Kaspersky documented Lazarus Operation FlashHole abusing CrossWeb EX with post-quantum encryption in South Korea | Source |
| 2026-10-22 | Kaspersky revealed APT41 BlackX UEFI bootkit exploiting Baton Drop on patched Windows in Central Asia | Source |
Device Code Phishing Kits Target Microsoft 365 at Scale
LevelBlue documented Wazza, a device code phishing kit active since July 2026 with 19 configurable themes that routes victims through Google Sites, .EU domains, and Cloudflare Workers before presenting a device code disguised as a document verification code. Backend sessions poll Microsoft authentication status every two seconds, capturing OAuth tokens when victims complete sign-in (LevelBlue. ANY.RUN reported SkyW4lker, a device code phishing kit sharing JavaScript modules, URL patterns, and hosting infrastructure with the Gabagool kit since April 2026, with 618 sandbox analyses identified between June and October 2026 growing from 20 to 200 monthly submissions (ANY.RUN.
LevelBlue separately reported BlackApex, DarkApex, SPY, and XXX101 phishing kits abusing Microsoft 365 device code authentication to steal access tokens, refresh tokens, and Primary Refresh Tokens, with over 70 panels sharing one backend with AI-generated code. PRT compromise enables SSO persistence across Microsoft 365 applications comparable to Golden Ticket attacks (LevelBlue. Huntress documented a ClickFix variant targeting macOS users via malicious sponsored Google results leading to fake custom GPT pages that issue Terminal commands to steal Microsoft 365 session cookies, with attackers creating hidden inbox rules to intercept invoices and redirect payments (Huntress. Cofense reported a separate device code phishing campaign using fake construction bid invitations linking to spoofed Syncfusion and Mimecast pages that generate device codes for Microsoft's legitimate authentication portal (Cofense.
Sources: 1, 2, 3, 4, 5, 6, 7, 8
Infostealers Expand Targeting to AI Coding Assistant Credentials
Symantec reported Aoki (aka Aokigahara), a multi-functional infostealer first seen in early 2026 that added harvesting of AI coding assistant credentials including Claude, Codex, Cursor, Codeium, and Gemini. Stolen AI tokens enable attackers to impersonate developers, access connected repositories, cloud services, and CI/CD pipelines. The stealer also targets 50+ browsers, 70+ crypto wallet extensions, desktop wallets, and password managers (Symantec. Check Point published parallel analysis confirming Aokigahara targets Claude, ChatGPT Desktop, Cursor, Windsurf, and GitHub Copilot session tokens and API keys, with C2 using X25519 key exchange and AES-GCM encryption (Check Point.
Malpedia documented Warden Stealer, a Rust-based MaaS infostealer active since May 2026 with approximately 110 operators, which added built-in collection of tokens from AI coding agents including Claude, Codex, and Cursor in version 1.9. It bypasses Chromium App-Bound Encryption by scanning browser memory for v20_master_key and injecting shellcode to call CryptUnprotectMemory (Malpedia. Talos reported VOLTA infostealer enriching stolen data via Gmail receipt searches and food delivery APIs, then POSTing structured bundles to OpenRouter for AI-driven victim profiling using Claude Opus 4.6 as the default model, with fallback to Kimi K2 and Gemini 2.5 Flash (Talos.
Reporting chronology
| Publication date | Reported update | Source |
|---|---|---|
| 2026-10-08 | Malpedia documented Warden Stealer adding AI coding agent token collection in version 1.9 | Source |
| 2026-10-15 | Symantec reported Aoki infostealer targeting Claude, Codex, Cursor, and Gemini credentials | Source |
| 2026-10-23 | Check Point confirmed Aokigahara targeting AI app session data with X25519-AES-GCM C2 | Source |
| 2026-10-31 | Talos reported VOLTA using OpenRouter AI for victim profiling with Claude Opus 4.6 default model | Source |
eMudhra Weak Intermediate CA Certificates Enable Forgery
Talos reported that a flawed key generation ceremony on April 30, 2025 produced eight eMudhra intermediate CA certificates with P-384 elliptic curve keys containing only 3-8 bits of entropy instead of 384. An attacker recovering these private keys can forge TLS or code-signing certificates trusted by clients that have not blocked them. Six compromised TLS certificates were blocked by Chrome, Mozilla, and Apple, and two code-signing certificates were blocked by Microsoft. Go assigned CVE-2026-111943 and released versions 1.27.3 and 1.26.10 that filter the compromised SPKI hashes (Talos.
No malicious exploitation was observed as of October 21, 2026. eMudhra revoked the compromised certificates and stopped issuing new certificates under its G3 root. The weakness originated from a wrap/unwrap flaw during the key generation ceremony for 24 subordinate CAs. Akamai separately reported that attackers compromised three country-code top-level domain registries (.gh, .sl, .as) between September 22-27, 2026, obtaining 12 TLS certificates for Google and YouTube domains via DNS manipulation, with all certificates revoked by October 1 (Akamai.
ICS/OT Vulnerability Disclosures Across Energy and Manufacturing
CISA issued multiple ICS advisories during October 2026 across energy, manufacturing, transportation, and healthcare sectors. Armatura One physical access-control systems were disclosed with CVE-2023-46604 in an embedded Apache ActiveMQ enabling unauthenticated RCE, plus hard-coded AES keys and fixed database passwords identical across all installations (CISA. Red Lion Controls N-Tron 700 Series industrial switches had seven CVEs including default factory credentials that persist after configuring new accounts and unauthenticated SNMP access enabling config exfiltration and firmware changes (CISA. Grid Protection Alliance openPDC and openHistorian software had a CVSS 9.8 deserialization flaw enabling unauthenticated RCE via the service console, with Docker images shipping fixed admin credentials and no planned fix (CISA.
Siemens disclosed five vulnerabilities in APOGEE, TALON, and Desigo building automation controllers exploitable by unauthenticated remote attackers via web or FTP services, with no fix planned for some products (CISA. Hitachi Energy RTU500 series had nine CVEs in OpenSSL, libxml2, and libcurl components enabling DoS, credential leakage, and mTLS bypass (CISA. EPICS Base releases through 7.0.10 had four network protocol parsing vulnerabilities including a CVSS 9.8 flaw potentially enabling unauthenticated RCE in critical manufacturing environments (CISA. Mitsubishi Electric disclosed CVE-2026-16436 enabling remote authentication bypass via a fixed challenge response shared across all users across ten industrial monitoring products, with MC Works64, GENESIS32, and BizViz having no planned fixes (CISA.
Sources: 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19, 20, 21
Other notable reporting
- Huntress Tragic Quadrant: Threats Wrecking Small and Mid-Sized Businesses: Huntress analyzed threats across 5 million endpoints and 300,000 organizations, finding RMM tool abuse in 45% of endpoint incidents and mailbox manipulation driving 24.6% of identity threats. EvilTokens PhaaS hit 344 organizations across 5 countries in 16 days using device code phishing.
- Zscaler ThreatLabz 2026 Ransomware Report: Zscaler reported top 10 ransomware groups exfiltrated 896.2 TB of data, a 275.8% year-over-year increase, with 62% of victims holding manager-level titles. Attackers target non-IT staff through Microsoft Teams vishing combined with Quick Assist for remote access.
- Japan Adopts Proactive Cyber Defense Strategy: Recorded Future analyzed Japan's shift from voluntary information sharing to mandatory incident reporting starting October 2026, with broader monitoring powers in November 2027. Japan recorded 686.2 billion cyberattack-related packets in 2024, with 99.4% using overseas source IPs. TraderTraitor stole approximately $308M from DMM Bitcoin.
- Russia Tightening Internet Controls Creates Operational Risks: Recorded Future documented Russia blocking 460 VPN services by early 2026, conducting 37,000+ hours of managed internet blackouts across 80% of territory, and expanding SORM interception to commercial banks. FSB installed Monokle spyware on a detained analyst, setting precedent for domestic spyware use.
- Trail of Bits: Package Managers Have an Integrity Problem: Testing 25 package managers found fewer than half verify package integrity by default and only two verify every downloaded package. Several compute hashes but install mismatched packages anyway. Only Go verifies signatures by default using a checksum database and transparency log.
- Project Zero: How to Fix a Bug in a Fix: Google Project Zero analyzed four emergency patching methods: feature flags, input filtering, alternate update channels, and hotpatching. LLM-accelerated vulnerability discovery makes rapid remediation capability urgent for vendors, as testing and delivery stages remain the primary bottlenecks.
- ESET APT Activity Report Q2-Q3 2026: ESET documented Belarus-aligned FrostyNeighbor conducting the first destructive cyberattack in the European theater against Ukrainian manufacturing, DeceptiveDevelopment shifting to supply-chain attacks, and multiple APT groups leveraging AI/LLMs for malware development and social engineering.
- CrowdStrike 2026 Legal Threat Landscape Report: Cyberattacks on legal entities rose 39% year-over-year with 455 named on leak sites. CHATTY SPIDER led eCrime activity with 129% surge. China-nexus WARP PANDA exploited CVE-2026-41709 in VMware ESXi as a zero-day, and TITAN PANDA targeted Intapp Time Server billing software.
Vulnerability round-up
Endpoint software
| CVE | Affected product | Reported development | Sources |
|---|---|---|---|
CVE-2026-109618 | PHP core (session.upload_progress, multipart/form-data handling) | Disclosure of CVSS 9.2 unauthenticated RCE via use-after-free in OOM handling; PoC released same day; CISA KEV addition October 26; exploitation confirmed on Windows IIS | 1, 2, 3, 4 |
CVE-2026-109619 | PHP core (session.upload_progress, multipart/form-data handling) | Disclosure of CVSS 9.2 unauthenticated RCE via stale internal state reuse across worker requests; PoC released; CISA KEV addition October 26 | 1, 2, 3 |
CVE-2026-111497 | Google Brotli compression library (c/dec/decode.c) | Disclosure of CVSS 9.2 OOB write in Brotli decoder affecting Node.js, Bun, .NET, Envoy, Python; CISA KEV addition October 27; active exploitation confirmed | 1, 2, 3 |
CVE-2026-96951 | Microsoft Windows Shell | Security feature bypass via command injection; actively exploited in the wild; CISA KEV listed; patched in October 2026 Patch Tuesday | 1, 2 |
CVE-2026-111943 | Go standard library crypto/x509 | Go X.509 chain verification does not filter compromised eMudhra intermediate certificates with 3-8 bits of entropy; fixed in Go 1.27.3 and 1.26.10 | 1 |
CVE-2026-82077 | PaperCut NG (Scan-to-Fax providerConnector field) | Post-auth RCE via relative path traversal and command-args injection; disclosed by WatchTowr with full chain | 1 |
CVE-2026-81578 | PaperCut NG (Apache Tapestry 3 service parameter handling) | Authentication bypass via service parameter page chaining; exploited by AI-assisted actors against 440 instances | 1, 2 |
CVE-2026-46300 | Linux Kernel | Missing initialization of a variable; CISA KEV addition October 28; actively exploited | 1 |
Security appliances
| CVE | Affected product | Reported development | Sources |
|---|---|---|---|
CVE-2026-104935 | Ivanti Connect Secure VPN | Zero-day authentication bypass and RCE via welcome.cgi and wu.cgi race condition; CISA KEV addition October 13; active exploitation by UTA0576 confirmed | 1, 2 |
CVE-2026-111357 | Ivanti Connect Secure | Missing authentication for critical function; CISA KEV addition October 30; actively exploited | 1 |
CVE-2026-118527 | Ivanti Connect Secure | Predictable identifiers or tokens enabling session forgery; CISA KEV addition October 30; actively exploited | 1 |
CVE-2026-21589 | Multiple Atlassian Data Center products (atlassian-plugins-webresource library) | Pre-auth arbitrary file read via double-colon path traversal across eight products; actively exploited; CISA KEV listed | 1, 2, 3 |
CVE-2026-102764 | BeyondTrust Remote Support (formerly Bomgar) | Authentication bypass enabling unauthenticated file operations on self-hosted appliances; ARMORY CASTLE exploitation; CISA KEV addition October 16 | 1, 2 |
CVE-2026-88779 | Citrix NetScaler ADC and NetScaler Gateway (SAML-configured) | Memory overflow in SAML SP/IdP configurations causing DoS; CISA KEV addition October 4; actively exploited | 1, 2 |
CVE-2026-107406 | Citrix NetScaler ADC and NetScaler Gateway | RCE or DoS affecting SAML service provider or identity provider configurations; advisory issued October 9 | 1, 2 |
CVE-2026-76504 | Cisco Catalyst SD-WAN Manager | API authentication bypass enabling unauthenticated access to SD-WAN management plane; CISA KEV listed; actively exploited with no fix available initially | 1, 2 |
CVE-2026-76519 | Cisco Secure Email Gateway AMP / Content Scanner Tools | CVSS 9.8 relative path traversal via crafted email attachments enabling unauthenticated root file write and command execution; CISA KEV addition October 14; actively exploited | 1, 2 |
CVE-2026-0322 | Palo Alto Networks PAN-OS Management Interface | Authenticated RCE in management interface; CISA KEV addition October 14; actively exploited | 1, 2 |
CVE-2026-105133 | AhsayCBS checkSysPwd function | Authentication bypass via random token substitution; CISA KEV addition October 14; actively exploited for webshell and XMRig deployment | 1, 2 |
CVE-2026-105134 | AhsayCBS Replication Receiver (UpdateReceivers API) | Unauthenticated RCE as NT AUTHORITY/SYSTEM; CISA KEV addition October 14; actively exploited | 1, 2 |
CVE-2026-104286 | Fortinet FortiMail | Path traversal allowing access to restricted directories; CISA KEV addition October 1; actively exploited | 1 |
CVE-2026-20127 | Cisco Catalyst SD-WAN Controller, Manager, and Validator | CVSS 10.0 authentication bypass enabling admin access; exploited by UAT-11473 to deploy KeyDrift malware stealing AES-256 IPsec keys | 1, 2 |
CVE-2026-76418 | Cisco Secure Firewall ASA and FTD SSL VPN | Remote Access SSL VPN Tunnel denial-of-service; CISA KEV addition October 29; actively exploited | 1, 2 |
CVE-2026-114625 | Arista VeloCloud Orchestrator | SQL injection allowing database manipulation and system compromise; CISA KEV addition October 23; actively exploited | 1, 2 |
CVE-2026-41709 | Broadcom VMware ESX | Insufficient logging vulnerability; CISA KEV addition October 28; actively exploited by WARP PANDA as zero-day | 1, 2 |
CVE-2026-106173 | Fortinet FortiAnalyzer, FortiManager, FortiMail, FortiSIEM, FortiSandbox | SQL injection and arbitrary file write; actively exploited in the wild | 1 |
Cloud and developer tooling
| CVE | Affected product | Reported development | Sources |
|---|---|---|---|
CVE-2026-97229 | GitLab CE/EE CI/CD regex compiler | CVSS 10.0 unauthenticated RCE via integer overflow in regex validation; CISA KEV addition October 29; actively exploited | 1, 2 |
CVE-2026-110264 | GitLab Community and Enterprise Editions | CVSS 10.0 integer overflow in regex compiler enabling unauthenticated RCE; affects all branches 7.0 through 19.5 | 1 |
CVE-2026-102489 | Zammad customer support platform | Session fixation enabling session hijacking; CISA KEV addition October 2; actively exploited | 1 |
CVE-2026-102490 | Zammad customer support platform | Improper privilege management enabling unauthorized escalation; CISA KEV addition October 2; actively exploited | 1 |
CVE-2026-35029 | LiteLLM gateway | Missing RBAC on /config/update enables config tampering, file read, and RCE via S3-hosted Python modules; fixed in v1.83.9 | 1 |
CVE-2026-59823 | LiteLLM gateway | SSRF via user_config parameter allowing LLM provider API key exfiltration; fixed in v1.83.9 | 1 |
CVE-2026-110447 | Collabora Online Built-in CODE Server | OS command injection enabling total control post-exploitation; CISA KEV addition October 23; actively exploited | 1 |
CVE-2026-48356 | Adobe Commerce and Magento | Unrestricted upload of dangerous file types enabling code execution; CISA KEV addition October 27; actively exploited | 1 |
CVE-2026-25243 | Redis Software / Redis OSS/CE | Heap memory corruption in RESTORE command; actively exploited per open-source reporting | 1, 2 |
CVE-2026-118539 | Drupal Commerce Core | Access bypass vulnerability; actively exploited in the wild | 1 |
Industrial systems
| CVE | Affected product | Reported development | Sources |
|---|---|---|---|
CVE-2026-100730 | Grid Protection Alliance openPDC and openHistorian | CVSS 9.8 deserialization of untrusted data enabling unauthenticated RCE via service console; Docker image ships with fixed admin credentials | 1 |
CVE-2026-32645 | Red Lion Controls N-Tron 700 Series | Default factory admin credentials persist after configuring new accounts; critical for industrial network switch security | 1 |
CVE-2026-114595 | EPICS Base | CVSS 9.8 integer overflow in network request processing enabling potential unauthenticated RCE in critical manufacturing | 1 |
Coverage
The source collection contained 280 articles.
Report classifications: Critical: 70; High: 163; Medium: 38; Low: 9.
Coverage by category: vulnerability: 113; malware: 51; threat actor: 34; general security news: 31; APT: 24; phishing/social engineering: 24; data breach: 3.
Counts describe collected reporting, not incidents or affected organizations.