Ivanti Connect Secure Zero-Day Exploitation, ARTEX AI Penetration Framework, and October Patch Tuesday
The week's reporting centered on active exploitation of a zero-day chain in Ivanti Connect Secure VPN appliances, with Volexity attributing the activity to a previously unknown actor assessed with medium confidence to use LLMs in exploit development. CISA added the primary vulnerability to its Known Exploited Vulnerabilities catalog and warned that the vendor's integrity checker may not detect compromise. Separately, multiple research teams documented widespread abuse of ARTEX, a Chinese-language open-source AI penetration testing framework, finding over 600 exposed instances and campaigns against South Korean financial institutions that exfiltrated personal data on tens of thousands of individuals. Other major developments included Microsoft's October Patch Tuesday covering 1,415 vulnerabilities with one already exploited in the wild, a coordinated disruption of Chinese-language phishing kits that delivered over one billion malicious messages since January, Mandiant's disclosure of ARMORY CASTLE exploiting BeyondTrust Remote Support, and parallel investigations by Talos and Recorded Future into a Kazakhstan-nexus espionage group targeting CIS governments. Unit 42 detailed both a memory-resident FortiGate implant called PEERSTITCH and an AI-assisted campaign deploying SysGoRAT against government Linux infrastructure.
Improve this report
Send corrections, useful missing information, or quality feedback. It is private and will not appear as a comment.
Detection / HunterZ.ai
Main developments
Ivanti Connect Secure Zero-Day Chain Exploitation
Volexity published a technical analysis of chained exploitation of previously unknown vulnerabilities in Ivanti Connect Secure VPN appliances on October 13, 2026. The researchers found that attackers bypassed authentication and achieved remote code execution by exploiting a race condition in the welcome.cgi and wu.cgi components, then stole credentials including administrator recovery keys and performed Active Directory reconnaissance. Volexity attributed the activity to a tracked actor designated UTA0576 with medium confidence, noting that the actor appeared to use large language models for exploit development and post-exploitation work. A separate actor, UTA0533, was noted using a similar LDAP packet capture technique against SonicWall SMA VPN appliances. Ivanti released a partial fix for CVE-2026-104935 on October 13, but Volexity reported that a privilege escalation component remained unpatched, and that all commands and results passed through exploit channel HTTPS responses without deploying a web shell, making the Ivanti Integrity Checker unable to detect exploitation because library patches are memory-only and restored. CISA issued a same-day alert confirming active exploitation, noting affected versions including 22.8R2.4 and prior, 25.1.1 and prior, and 25.1.2, and warning that version 22.8 reaches end of life on January 31, 2027. Talos's separate Q3 vulnerability assessment also covered the Ivanti flaw, noting it enables unauthenticated RCE with elevated privileges on fully patched VPN. The Canadian Cyber Centre added both CVE-2026-104935 and a separate Ivanti vulnerability CVE-2026-111357 to its October 16 advisory digest, the latter reported exploited in the wild per open-source reporting.
ARTEX AI Penetration Testing Framework Abuse
Multiple research teams reported on ARTEX, a Chinese-language open-source AI penetration testing framework that uses multi-agent large language model orchestration for scanning, exploitation, and lateral movement. AhnLab's ASEC found over 600 ARTEX instances exposed on public IPs with approximately 900 associated IPs, two of which contained operational directories with web shells, pass-the-hash tools, Kubernetes scan results, and implants. In laboratory testing, ARTEX autonomously exploited CVE-2023-46604 against Apache ActiveMQ and spread to four internal hosts without human guidance. ASEC also documented the ARTEX toolchain being rebranded and copied by multiple operators including Bob Research Labs, which used LLM profiles based on DeepSeek, GLM, Kimi, and Claude to exploit CVE-2026-75650 in Magento and deploy a remote-control bot via IRC. Zscaler separately reported that a Chinese-speaking financially motivated actor used ARTEX against South Korean financial institutions starting late September 2026, exfiltrating PII of approximately 67,000 individuals, with ARTEX guardrails blocking .gov and .edu targets but not financial institutions. Recorded Future attributed similar activity to TAG-217, reporting at least 68,000 individuals' personal data exposed across at least seven financial institutions, with the actor also using Claude Code, DeepSeek, Grok, and Zhipu AI. Check Point's weekly intelligence report also noted the ARTEX toolkit campaign against South Korean financial institutions using Claude Code AI sessions.
Microsoft October 2026 Patch Tuesday
Talos reported that Microsoft's October 2026 Patch Tuesday addressed 1,415 vulnerabilities across its products, including 85 rated critical. CVE-2026-96951, a Windows Shell command injection vulnerability with CVSS 8.8, was confirmed as exploited in the wild and added to the CISA Known Exploited Vulnerabilities catalog. Notable among the critical fixes were CVE-2026-96207 in Microsoft Partner Center rated at CVSS 10.0 for elevation of privilege via improper certificate validation, five Outlook remote code execution vulnerabilities rated 9.8, and CVE-2026-100202 in Windows Message Queuing also rated 9.8. Talos released corresponding Snort 2 rules (1:67210 through 1:67322) and Snort 3 rules (1:301663 through 1:301718). The Canadian Cyber Centre confirmed in its October 14 advisory digest that CVE-2026-96951 was actively exploited, affecting Microsoft Exchange Server 2016, 2019, and Subscription Edition RTM, alongside Adobe ColdFusion and Photoshop updates.
Chinese-Language Phishing Kit Disruption (CoGUI, YAM Store, Iron Man)
Microsoft published findings on a Chinese-language cybercrime network operating through an underground marketplace called YAM Store, which sold phishing kits and stolen cloud accounts. Three kits tracked under Storm-2990 — CoGUI, Haozi, and Iron Man System — collectively enabled over one billion phishing emails and text messages targeting more than 75,000 organizations since January 2026 across healthcare, finance, and technology sectors. Iron Man System alone contained 2,335 phishing templates targeting 139 countries. A separate tracked entity, Storm-3119, sold compromised Microsoft 365 and Azure Education tenants using app-only authentication for durable persistence. Proofpoint simultaneously published on a disruption targeting three related phishing kits — CoGUI, CoAceV, and CoFish_Ws — that shared a common URL pattern embedding kumamoto-city.jp in the path and used decoy redirects to evade sandbox analysis. Proofpoint data showed CoGUI alone accounted for 1,081 campaigns across 13,586 unique domains. Microsoft and Proofpoint both noted that law enforcement partners seized many domains used in these attacks. Japanese authorities reportedly linked similar phishing activity to financial gains used to purchase Chinese stocks.
ARMORY CASTLE Exploits BeyondTrust Remote Support
Mandiant disclosed that ARMORY CASTLE, a suspected PRC-nexus espionage actor formerly tracked as UNC5221, exploited CVE-2026-102764 in BeyondTrust Remote Support to gain unauthenticated initial access to self-hosted appliances. The vulnerability allows arbitrary file operations on the appliance without authentication, affecting all supported Remote Support versions up to 26.2.3. Mandiant documented custom tooling including SAWFLY, a C2 using Base32 transport with 256-bit RC4 encryption, and BARECLAW, a C-based SOCKS5 proxy tunneler over WebSocket using CivetWeb. Persistence was achieved through sudoers template modifications surviving reboots and configuration resets. The BT26-05 patch was applied to all SaaS tenants on October 9, 2026. CISA issued a separate alert on October 16 confirming active exploitation, noting that the flaw enables threat actors to gain initial access, establish persistence, escalate privileges, and move laterally, and adding it to the KEV catalog the same day. The Canadian Cyber Centre also reported the vulnerability as exploited in the wild in its October 16 digest.
UAT-6275 / TAG-193: Kazakhstan-Nexus Espionage Against CIS Governments
Cisco Talos and Recorded Future published parallel investigations into a Kazakhstan-nexus espionage group designated UAT-6275 by Talos and TAG-193 by Recorded Future, with overlaps to the Silent Lynx and YoroTrooper tracking names. The group has targeted government agencies across Russia, Tajikistan, Belarus, Armenia, Uzbekistan, Azerbaijan, and Turkmenistan since early 2025, focusing on diplomacy, water resources, and energy information. Talos identified four infection chains that evolved over 14 months from VBA macros to exploitation of CVE-2025-8088 in WinRAR, which uses alternate data stream path traversal to drop payloads in the Windows Startup folder. Custom reverse shells named Chvost, Laplas, and PickleShell provided interactive remote access, with an exposed GitHub repository hosting PowerShell scripts with trackable commit history. Operator working hours aligned with UTC+5 Astana time, supporting the Kazakhstan nexus assessment. Recorded Future's separate reporting noted the same toolkit including LAPLAS and multiple FoalShell variants, and documented that the group abuses GitHub to host payloads alongside compromised government email accounts. Talos also reported the recovery of a FortiWeb CVE-2025-25257 proof-of-concept from the group's staging directories.
PEERSTITCH: Memory-Resident FortiGate Implant
Unit 42 disclosed PEERSTITCH, a memory-resident ARM64 ptrace injector targeting FortiGate devices, deployed after exploitation of CVE-2024-47575 (FortiJump) in FortiManager. The implant operates as a one-shot injector targeting the /bin/miglogd and /bin/iked processes with no disk persistence, making it invisible to standard security tools. The campaign compromised a central management device at a telecommunications company, enabling the attackers to exfiltrate configuration files, credentials, and encryption keys from over 500 managed firewalls. Attackers impersonated FortiManager using stolen trust material to pull configurations from managed firewalls over TCP port 541. Unit 72 also documented abuse of Apache mod_md ACME DNS-01 challenge functionality, mapping it to /bin/node for a persistent reverse shell command-and-control channel, and noted anti-forensics including killing miglogd and syslogd, clearing crash logs, deleting staging files, and rolling back rogue configurations. The tracking cluster was designated CL-STA-0954, with UNC5820 as an alternative designation.
Sources: 1
AI-Powered Espionage Campaign Targeting Government Linux Infrastructure
Unit 72 attributed a campaign active from October 2025 through April 2026 to a tracked cluster designated CL-UNK-1211, finding that attackers used AI to help generate payloads targeting government Linux servers running PHP, Plesk, and mail services. The campaign deployed SysGoRAT, a Go-based cross-platform remote access trojan with seven command types and six persistence methods, defaulting its C2 to a WebSocket connection on port 443. Persistence was achieved by modifying PHP auto_prepend_file directives across /etc/php/ and /opt/plesk/php/ to load a malicious driver file before any PHP script execution. Credential theft was performed by hijacking the .bashrc sudo alias to write base64-encoded passwords to a disguised log file. Attackers also deployed a custom Go scanner for CVE-2025-55182 and used PwnKit (CVE-2021-4034) for privilege escalation. Data exfiltration blended with legitimate traffic through transfer.sh, bashupload.com, filebin.net, and files.catbox.moe. Anti-forensics included log deletion, timestamp modification, and disguising activity as security testing.
Sources: 1
Ephemeral Supply Chain Tampering of AOMEI Update Infrastructure
Mandiant reported that attackers designated UNC7017 briefly modified AOMEI software's update servers in 5-to-10-minute windows, causing some users to download malware disguised as legitimate updates during those intervals. The brevity of tampering windows evaded both point-in-time triage and automated scanning. Mandiant documented two related tools: CROSSROAD, a multi-stage downloader that checks for AOMEI substring in the file path and aborts if geolocation returns China, and SHINYROAD, a modular backdoor using an embedded PocketPy interpreter for in-memory Python bytecode execution. A third tool, ADAPTAGENT, functioned as an Adaptix C2 client masquerading as Nitro PDF tools. Mandiant noted that server-side remediation by AOMEI does not disinfect endpoints compromised during tampering windows, meaning affected systems may retain backdoor installations and scheduled tasks from the April-through-July 2026 period.
Sources: 1
ConnectWise (SkyKick) Vendor Credentials Abused in Microsoft 365 Attacks
Huntress reported that attackers stole credentials for ConnectWise's Cloud Backup service, formerly SkyKick, and used them from a hosting provider's IP address beginning October 15, 2026 to access over 162 customer Microsoft 365 environments. The unauthorized access potentially exposed emails, files, and user directory information across affected tenants. Huntress also documented password spraying from the same IP address targeting four law firms on October 13 and 14 using curl/7.88.1, and found that the attacker added client secrets expiring in 2028 to MSP management and security tool applications. In at least one tenant, the ConnectWise Cloud Management Microsoft Teams app granted itself Global Administrator privileges. Huntress noted that directory reads via Microsoft Graph leave no audit trail, complicating detection.
Sources: 1
September 2026 CVE Landscape Analysis
Recorded Future's September 2026 CVE landscape report identified 94 high-severity vulnerabilities, a 29% increase from the prior month, with 47 enabling remote code execution across network-edge and security management products. Public proof-of-concept exploits existed for 75 of the 94. The analysis documented multiple exploitation campaigns including NightEagle deploying the GhostContainer backdoor on Exchange servers via compromised VPN credentials and BlueKeep, UNC3569 exploiting Sogou Input Method to deploy GRAYRABBIT, and UAT-12197, UAT-11823, and UAT-11988 each exploiting Cisco FMC for different post-exploitation objectives including root access, Cyclops Blink deployment, and Qilin ransomware. A China-linked actor designated Red Heron was documented exploiting Gitea, WordPress, and Proxmox VE vulnerabilities against 49 organizations. The report also noted that a Russian-speaking actor used AI agents to exploit PaperCut NG/MF vulnerabilities across 440 instances in 395 organizations, and that eight identified vulnerabilities were at least five years old, with the oldest approximately 12 years old. CWE-306 (Missing Authentication) was the most common weakness class.
Sources: 1
NetScaler CVE-2026-88771 Exploitation Analysis
Zscaler's ThreatLabz published a technical analysis of CVE-2026-88771 exploitation in Citrix NetScaler ADC and Gateway devices, a critical unauthenticated RCE with CVSS 9.8 that requires no optional features enabled. The attack stashes malicious code in web traffic logs and tricks a diagnostic script into executing it, using ${IFS} to avoid spaces and bypass request processing filters. Zscaler observed 28 source IPs in burst exploitation over six to seven minutes, deploying web shells named Nx_7508 across six directories with .php, .pl, and .html extensions. The analysis identified a shared download server at 209.141.43.95 also used by StyleSmuggler, the public proof-of-concept for CVE-2026-75650 in Magento, suggesting either a shared attacker or a common malware delivery service.
Sources: 1
Phishing and Social Engineering Campaigns
Huntress documented a fake WatchGuard VPN download site at watchguard-vpn.net, re-registered August 22, 2026 after a prior January takedown, that appeared in Bing search results and harvested VPN server addresses, usernames, and passwords. The site rejected the first credential submission to force a corrected second entry, then delivered the genuine signed WatchGuard installer from cdn.watchguard.com to avoid suspicion. Huntress attributed the wider campaign to Storm-2561, a financially motivated actor active since May 2025. In a separate report, Huntress detailed a ClickFix variant targeting macOS users via sponsored search results mimicking ChatGPT, where victims paste a Terminal command that steals Microsoft 365 login cookies, bypassing MFA entirely. The attacker then creates inbox rules named '..' to archive messages containing invoice numbers, enabling business email compromise within approximately 24 hours. Huntress also published on passkey-related threats, documenting the EvilTokens device code phishing-as-a-service kit that hit 344 organizations across five countries in approximately 16 days using AI-generated lures tailored per victim role. ANY.RUN separately reported on fake AI tool campaigns including counterfeit Claude, ChatGPT, and Copilot downloads, with 1,258 sandbox analyses between April 15 and October 5, 2026, finding US organizations among the top targets. Cofense documented a ScreenConnect RAT phishing campaign using fake event invitation emails and a legitimate signed ScreenConnect installer to bypass antivirus, connecting to an attacker relay server. Infoblox published on Hotejka, a Russian-language affiliate program operating a traffic distribution system that creates approximately 300 new Cloudflare Pages domains daily to deliver fake Microsoft security warnings and tech support scam content, rebranded from the ROI777 operation exposed in 2018.
Ransomware, Dark Web Activity, and FAMOUS CHOLLIMA Disruption
AhnLab's ASEC weekly ransomware summary reported that the n0n ransomware group attacked 12 South Korean port and logistics companies, a cyberattack targeted refrigeration systems at major French supermarket chains, and a Saudi fintech company suffered a claimed payment and transaction data leak. Detailed indicators and analysis were gated behind a subscription. Separately, CrowdStrike reported that the FBI, with CrowdStrike assistance, disrupted several U.S. laptop farms used by North Korean operatives in the FAMOUS CHOLLIMA campaign. The operatives obtain employment at American companies using fake identities, then redirect employer-issued laptops to these farms where associates install remote access software, enabling work from North Korea while appearing to be in the U.S. CrowdStrike noted that since May 2026, device hashes are spoofed using video capture cards disguised as webcams, and the same spoofed device has been reused across multiple victim organizations in aerospace, energy, finance, healthcare, defense, and government technology sectors.
CISA Known Exploited Vulnerabilities Catalog Additions
CISA made multiple additions to its Known Exploited Vulnerabilities catalog during the week. On October 14, CISA added four CVEs: CVE-2026-0322 affecting Palo Alto Networks PAN-OS via code injection, CVE-2026-76519 affecting Cisco Secure Email Gateway via relative path traversal, and CVE-2026-105133 and CVE-2026-105134 both affecting Ahsay AhsayCBS via improper authentication and command injection respectively. On October 16, CISA added two more: CVE-2025-61624, a path traversal vulnerability affecting multiple Fortinet products, and CVE-2026-102764 in BeyondTrust Remote Support, an authentication bypass by spoofing. The Canadian Cyber Centre's October 15 advisory digest separately confirmed CVE-2026-0322 as actively exploited in the PAN-OS management interface.
Midnight Blizzard-Linked Campaign Traced from File Hashes
ReversingLabs published an investigation tracing a Midnight Blizzard-linked campaign from two file hashes to a network of phishing domains and servers targeting government and think-tank organizations in Ukraine and Europe using fake Microsoft 365 login pages. The researchers identified five rebuilt stagers with different hashes sharing one 35-item capability profile and identical beacon labels, and found that a staging server at 213.145.86.112 still served malicious stagers on October 5, eight weeks after Microsoft's prior public report. A Go backdoor communicating with 192.142.52.31, an address absent from both vendor reports, was also identified. The campaign used AI tools to repeatedly rebuild malware, and delivery shortcuts were 60-byte .url files that AV consensus rated as goodware.
Sources: 1
Aoki Infostealer Targeting AI Coding Assistant Credentials
Symantec reported on Aoki, an infostealer first observed in early 2026 that recently added AI credential theft capabilities. The malware targets .claude.json, Codex config.toml, .env files, and session cookies for multiple AI coding tools, while also stealing data from over 50 browsers, 70-plus crypto wallet extensions, desktop wallets, and password managers. Aoki also records installed security products on victim machines for reconnaissance. Older samples used unencrypted HTTP on port 8080, while newer samples use domains with HTTPS gate endpoints. Symantec detects this family as Infostealer.Aoki, and the threat group behind it remains unidentified.
Sources: 1
LiteLLM Gateway Vulnerabilities
Check Point Research disclosed vulnerabilities in LiteLLM, an open-source gateway for AI language model APIs. CVE-2026-35029 involves missing role-based access control on the /config/update endpoint, enabling config tampering, file read, and remote code execution through prefix matching on /config in the UI route allowlist that bypasses stricter API route checks. CVE-2026-59823 is an SSRF vulnerability via the user_config parameter in /chat/completions that bypasses configurable clientside auth parameter restrictions to enable LLM provider API key exfiltration. The research also found that password hashes use single-iteration unsalted hashing, exposed to low-privilege users via UI tokens. Both vulnerabilities were fixed in LiteLLM version 1.83.9.
Sources: 1
Medical and IoT Device Vulnerability Disclosures
CISA published multiple ICS and medical advisories during the week. For the J4Care dcm4chee-arc-light medical imaging system, six vulnerabilities were disclosed in version 5.35.1, including CVE-2026-105279 allowing unauthenticated DICOM operations to enumerate, retrieve, and store patient data, and CVE-2026-105667 enabling unauthenticated HL7 messages to irreversibly reassign patient records. Three additional CVEs cause denial of service through parser infinite loops. For Thermo Fisher Torrent Suite Software versions 5.18 and older, CVE-2026-107777 allows an authenticated attacker to read arbitrary files via path traversal. CISA also disclosed vulnerabilities in Guangzhou Juan IoT Technology's EseeCloud cameras, including CVE-2026-107617 with CVSS 9.8 enabling unauthenticated login via hard-coded credentials, CVE-2026-97858 allowing device theft and rebinding, and CVE-2026-105664 exposing a device enumeration oracle; the vendor did not respond to CISA coordination. Swann SWIFI-SE2KIC security cameras were also affected by three local authentication bypass vulnerabilities (CVE-2026-107605 through CVE-2026-107607) with no patch available. Separately, CISA disclosed CVE-2026-92817 in Inductive Automation Ignition, an incorrect authorization flaw allowing authenticated Vision users to escalate to Designer-level access and execute arbitrary code.
September 2026 APT Group Threat Trends
AhnLab's ASEC published a September 2026 APT trend report documenting multiple active threat groups. Famous Chollima's Contagious Interview campaign delivered OtterCookie malware with GitHub personal access token theft. Fire Ant compromised Cisco IOS XR routers and TACACS authentication servers. NightEagle combined the GhostContainer Exchange backdoor with BlueKeep and Impacket for lateral movement. UAC-0099 used GuardBreaker, an LLM-based obfuscation technique designed to disrupt AI analysis. APT36 deployed RUSTYSHADE, RUSTYMOVE, PSNATCH, and BASHNATCH via GitHub private repositories and USB drives. The report also covered Kimsuky's segmented C2 using GitHub, WordPress, Dropbox, and Pinggy, Konni's VelvetCake downloader, and Transparent Tribe's multi-format attacks against government and defense targets.
Sources: 1
Russia Tightening Internet Controls
Recorded Future published research documenting that since mid-2025, the Russian government has blocked 460 VPN services by early 2026, a 70% increase from October 2025, using TSPU DPI hardware installed at major ISPs for protocol-level VPN blocking and traffic throttling. SORM interception hardware mandates were expanded to Russian commercial banks in early 2026. The FSB installed Monokle Android spyware, developed by the sanctioned Special Technology Center, on a detained analyst, setting a precedent for domestic spyware use against Russian nationals. By April 2026, 74 of 83 Russian regions had introduced allowlists of government-approved sites. The measures create operational risks including surveillance, data exposure, and disruption of business tools for organizations operating in Russia.
Sources: 1
GitLab Critical Vulnerability Disclosure
CERT-EU published Security Advisory 2026-016 covering 24 vulnerabilities fixed in GitLab's self-hosted CE and EE software. The most severe, CVE-2026-97229 with CVSS 10.0, enables unauthenticated arbitrary code execution via an integer overflow in the CI/CD regular expression compiler, triggered by sending a crafted regular expression through GitLab's CI/CD pipeline configuration. Fixed versions are 19.4.2, 19.3.4, and 19.2.8. CVE-2026-92883 with CVSS 8.7 enables authenticated cross-site scripting via crafted repository syntax highlighting, and CVE-2026-16301 allows developers to run pipelines on protected branches without proper authorization. The advisory noted 21 additional issues including cache poisoning rated 8.2 that can expose protected CI/CD variables.
Sources: 1
Cisco Secure Email Gateway Active Exploitation
CERT-EU published Security Advisory 2026-017 documenting active exploitation of CVE-2026-76519 in Cisco Secure Email Gateway's AMP and Content Scanner Tools component. The vulnerability, rated CVSS 9.8, is a relative path traversal via crafted email attachments that allows unauthenticated remote root file write and command execution. Cisco PSIRT confirmed active exploitation as of October 2026. The advisory noted that File Reputation service must be enabled and assigned to a mail policy for exposure, and that compromised appliances may expose private SSH keys used for cluster authentication. The fix is available in Content Scanner Tools version 25.3.0.5525 and later, with affected AsyncOS versions including 15.5 before 15.5.5-014, 16.0, and 16.5 before 16.5.0-780.
Sources: 1
Other notable reporting
- AgentTesla Distribution via Business Proposal Email: ASEC documented an AgentTesla distribution campaign using fake business proposal emails with ZIP attachments containing multi-stage .NET loaders that reconstruct subsequent DLLs from RGB pixel data in bitmap resources, masquerading as café management and security software.
- Sophos EDR Killer Analysis: Sophos published analysis of EDR killer techniques including the EDRKillShifter tool used in a 2024 RansomHub attack attempt, documenting how attackers load trusted but vulnerable drivers to disable endpoint protection, and how Sophos blocks these via cryptographic hash matching and separate kernel filter drivers.
- AI Governance and Unauthorized AI Tool Usage: Sophos reported that 82% of organizations have discovered unauthorized AI tools unknown to IT or security, and described a Vercel breach originating from a consumer AI product connected to a corporate Google Workspace account.
- Microsoft VSS Limitations and Ransomware: Huntress published a technical analysis of Microsoft Volume Shadow Copy Service limitations, noting the 64TB volume cap, write latency during backup windows, and that ransomware variants deliberately delete local VSS snapshots on execution.
- Package Manager Integrity Analysis: Trail of Bits tested 25 popular package managers and found that most do not verify downloaded package integrity by default, with only Go verifying signatures through a checksum database and transparency log. The research noted that cpanminus uses unencrypted HTTP for Perl packages.
- Snowflake-Managed PostgreSQL RCE via Feature Chaining: Varonis disclosed a method to chain five normal PostgreSQL features to achieve server-level code execution in Snowflake's managed PostgreSQL service, escalating from CREATEROLE to superuser and rebuilding blocked file-access functions. Snowflake remediated server-side before publication.
- Sandbox Analysis Insights From 74K Submissions: ANY.RUN analyzed over 74,000 suspicious files submitted by a bank over three months, finding that half were malicious, PowerShell scripts had an 88.2% malicious yield, and the studied organization faced more malware than phishing compared to the banking industry average.
- CISA Election Security Advisory: CISA published a general advisory warning that threat actors use email, SMS, audio, and video content to target 2026 U.S. midterm election officials and campaign staff. No specific IOCs, malware families, or threat actors were identified.
- CISA Super Intelligence Agent Incident Response Guidance: CISA published voluntary core practices for organizations that train and evaluate advanced AI systems called super intelligence agents, covering limiting agent access, monitoring activity, preserving evidence, and reporting incidents and near-misses to government agencies.
- Ransomware GenAI Impact Analysis: Zscaler published analysis of how generative AI is shrinking the time between initial access and ransomware deployment, enabling any motivated attacker to create custom social engineering lures, data discovery tools, and bespoke malware using free or stolen LLM accounts.
Vulnerabilities discussed
| CVE | Affected product | Reported development | Sources |
|---|---|---|---|
CVE-2026-104935 | Ivanti Connect Secure VPN | Disclosure of zero-day chain with confirmed active exploitation; CISA KEV addition on October 13 | 1, 2, 3 |
CVE-2026-111357 | Ivanti | Reported exploited in the wild per open-source reporting; Canadian Cyber Centre advisory | 1 |
CVE-2026-102764 | BeyondTrust Remote Support (formerly Bomgar) | Exploitation report by Mandiant confirming ARMORY CASTLE campaign; CISA KEV addition on October 16 | 1, 2, 3 |
CVE-2026-96951 | Windows Shell | Exploited in the wild; CISA KEV addition; covered in October Patch Tuesday | 1, 2 |
CVE-2026-88771 | Citrix NetScaler ADC and NetScaler Gateway | Investigation and exploitation analysis by Zscaler ThreatLabz | 1 |
CVE-2026-107406 | Citrix NetScaler ADC and NetScaler Gateway | Investigation by Zscaler ThreatLabz | 1 |
CVE-2026-97229 | GitLab CE/EE CI/CD regex compiler | Disclosure by CERT-EU; CVSS 10.0 unauthenticated arbitrary code execution | 1 |
CVE-2026-92883 | GitLab CE/EE syntax highlighting | Disclosure by CERT-EU; authenticated cross-site scripting | 1 |
CVE-2026-16301 | GitLab CE/EE pipeline authorization | Disclosure by CERT-EU; authorization flaw on protected branches | 1 |
CVE-2026-76519 | Cisco Secure Email Gateway AMP / Content Scanner Tools | Exploitation report by CERT-EU with Cisco PSIRT confirmation; CISA KEV addition on October 14 | 1, 2 |
CVE-2026-0322 | Palo Alto Networks PAN-OS | Exploitation report; CISA KEV addition on October 14; Canadian Cyber Centre confirmation | 1, 2 |
CVE-2025-61624 | Fortinet Multiple Products | CISA KEV addition on October 16; path traversal vulnerability | 1, 2 |
CVE-2026-105133 | Ahsay AhsayCBS | CISA KEV addition on October 14; improper authentication | 1 |
CVE-2026-105134 | Ahsay AhsayCBS | CISA KEV addition on October 14; command injection | 1 |
CVE-2024-47575 | FortiManager | Exploitation report by Unit 72 in PEERSTITCH implant analysis | 1 |
CVE-2025-55182 | React2Shell | Exploitation report by Unit 72 documenting custom Go scanner deployment | 1 |
CVE-2026-75650 | Magento/Adobe Commerce | Exploitation report by ASEC in Bob Research Labs ARTEX environment and Zscaler infrastructure analysis | 1, 2 |
CVE-2026-35029 | LiteLLM gateway | Disclosure by Check Point Research; missing RBAC enabling config tampering and RCE | 1 |
CVE-2026-59823 | LiteLLM gateway | Disclosure by Check Point Research; SSRF enabling API key exfiltration | 1 |
CVE-2026-92817 | Inductive Automation Ignition (Perspective module) | Disclosure by CISA; incorrect authorization enabling privilege escalation to Designer-level | 1 |
CVE-2026-107617 | Guangzhou Juan IoT Technology EseeCloud App and Firmware | Disclosure by CISA; CVSS 9.8 hard-coded credentials enabling unauthenticated login | 1 |
CVE-2026-97858 | Guangzhou Juan IoT Technology EseeCloud App and Firmware | Disclosure by CISA; device theft and rebinding to attacker account | 1 |
CVE-2026-105664 | Guangzhou Juan IoT Technology EseeCloud App and Firmware | Disclosure by CISA; device enumeration oracle at scale | 1 |
CVE-2026-107605 | Swann SWIFI-SE2KIC firmware | Disclosure by CISA; local unauthenticated video feed access; no patch available | 1 |
CVE-2026-107606 | Swann SWIFI-SE2KIC firmware | Disclosure by CISA; local unauthenticated snapshot access; no patch available | 1 |
CVE-2026-107607 | Swann SWIFI-SE2KIC firmware | Disclosure by CISA; local unauthenticated real-time stream access; no patch available | 1 |
CVE-2026-97851 | J4Care dcm4chee-arc-light | Disclosure by CISA; missing authorization enabling deletion of referenced DICOM objects | 1 |
CVE-2026-105279 | J4Care dcm4chee-arc-light | Disclosure by CISA; missing authentication enabling unregistered DICOM peers to enumerate and retrieve patient data | 1 |
CVE-2026-105667 | J4Care dcm4chee-arc-light | Disclosure by CISA; unauthenticated HL7 messages enabling irreversible patient record reassignment | 1 |
CVE-2026-107777 | Thermo Fisher Torrent Suite Software | Disclosure by CISA; path traversal enabling authenticated arbitrary file read | 1 |
CVE-2025-8088 | WinRAR | Exploitation report by Talos and Recorded Future in UAT-6275/TAG-193 espionage campaign | 1, 2 |
CVE-2026-96207 | Microsoft Partner Center | Disclosure in October Patch Tuesday; CVSS 10.0 elevation of privilege | 1 |
CVE-2026-109812 | Moxa AIG-302 Series | Disclosure by Canadian Cyber Centre | 1 |
CVE-2026-109813 | Moxa AIG-302 Series | Disclosure by Canadian Cyber Centre | 1 |
Coverage
The source collection contained 57 articles.
Report classifications: Critical: 15; High: 33; Medium: 6; Low: 3.
Coverage by category: vulnerability: 22; APT: 8; general security news: 8; phishing/social engineering: 8; threat actor: 7; malware: 4.
Counts describe collected reporting, not incidents or affected organizations.