Weekly Threat Digest 26.44 || Oct 26

PHP2Shell leads a week of exploited flaws in VeloCloud, Collabora, and Plex

Share

Two critical flaws in PHP, together called PHP2Shell, let unauthenticated attackers run code on any PHP installation serving web requests. Fixes shipped October 21 after details leaked a day early, and public exploit code is circulating. CISA added exploited flaws in Arista VeloCloud Orchestrator, Collabora's built-in CODE server for Nextcloud, Plex Media Server, and Siemens Software Center to its Known Exploited Vulnerabilities catalog.

Nozomi Networks details how Iran-linked Handala entered California Water Service in June through an exposed GPS platform, and CISA reports SafePay ransomware has hit at least 600 victims. Two stories from last week grew: Microsoft attributes the ConnectWise Cloud Backup compromise to a China-state actor it tracks as Storm-0940, which used an AI pentesting framework against more than 2,250 Microsoft 365 tenants, and Forescout counted nearly 13,000 internet-exposed refrigeration controllers after the French supermarket attacks.

Act Now

  • PHP2Shell (CVE-2026-109618, CVE-2026-109619): Two critical flaws let unauthenticated attackers run code on PHP servers that serve web requests, and public exploit code is circulating. If you run PHP: upgrade to 8.5.12, 8.4.27, 8.3.36, or 8.2.35 and restart every worker.
  • Arista VeloCloud Orchestrator (CVE-2026-114625): An exploited SQL injection lets any authenticated user, even read-only, gain superuser access on on-premises orchestrators. CISA added it to KEV on October 23. If you run VCO on premises: update to 5.2.3.18, 6.1.3.10, or 6.4.2.12.
  • Collabora built-in CODE server for Nextcloud (CVE-2026-110447): Attackers are planting web shells through this unauthenticated command injection, and CISA added it to KEV on October 23. If you run the Built-in CODE Server app: update to 26.4.303 or remove it; disabling it isn't enough.

Drafted with AI from public reporting; reviewed and edited by GPP Labs before publishing.


OT/ICS Advisories

Marker shows the highest CVSS v3 score in each advisory: 🔴 9.0 and up · 🟠 7.0–8.9 · 🟡 4.0–6.9

  • 🔴 Siemens SIMATIC S7-1500 CPU 1518(F)-4 PN/DP MFP and S7-1500 TM MFP · ICSA-26-295-11 and ICSA-26-295-16 · 428 and 377 CVEs · CVSS 10.0. Used across manufacturing, transportation, energy, healthcare, and government. Vulnerabilities in the additional GNU/Linux subsystem allow privilege escalation, denial of service, and code execution. No fix available. Limit access to the Linux subsystem's shell to trusted personnel and run only trusted applications. CISA · CISA
  • 🔴 Siemens RUGGEDCOM CROSSBOW Station Access Controller · ICSA-26-295-15 · 5 CVEs · CVSS 9.8. Used in energy and critical infrastructure. SQLite vulnerabilities allow remote attackers to execute arbitrary code, cause denial of service, or obtain sensitive information through memory corruption and buffer overflows. Update to version 6.0 or later. CISA
  • 🔴 Siemens SIMATIC S7 PLCs Web Server · ICSA-26-295-14 · 8 CVEs · CVSS 9.6. Used across manufacturing, transportation, energy, healthcare, and government. Code injection and cross-site scripting flaws on the web server's Trace page let an attacker who tricks an authorized user into importing a crafted trace file run code in that user's browser session and perform PLC operations with that user's rights. Affects more than 150 S7-1500, ET 200SP, and Drive Controller CPU models. No fix available. Disable the web server if not needed, restrict ports 80 and 443 to trusted addresses, and import only trusted trace files. CISA
  • 🔴 Siemens Industrial Edge Device Kit and Devices · ICSA-26-295-09 and ICSA-26-295-17 · 2 CVEs · CVSS 9.1. Used across manufacturing, transportation, energy, healthcare, and government. Path traversal vulnerabilities allow authenticated remote attackers to access sensitive internal information, facilitate unauthorized access, or delete data. Update to version 1.26.2-1, 1.27.1-1, or 1.28.0-6 or later. Limit network access to trusted parties. CISA · CISA
  • 🟠 Hitachi Energy RTU500 remote terminal unit · ICSA-26-293-05 · 9 CVEs · CVSS 8.8. Used in energy sector. Vulnerabilities in IEC 60870-5-104, OpenSSL, libxml2, and libcurl allow remote attackers to cause denial of service, leak credentials, or bypass authentication. Update to firmware 13.9.1 or later. CISA
  • 🟠 Siemens RUGGEDCOM CROSSBOW network management platform · ICSA-26-295-12 · 11 CVEs · CVSS 8.8. Used in energy and critical infrastructure. SQL injection, authorization bypass, and privilege escalation flaws allow authenticated attackers to execute arbitrary SQL commands, retrieve credentials, or cause denial of service. Update to version 6.0 or later. CISA
  • 🟠 Siemens SIMATIC WinCC SCADA system · ICSA-26-295-13 · CVE-2026-59241 · CVSS 8.8. Used across manufacturing, transportation, energy, healthcare, and government. Database privilege escalation allows authenticated remote attackers with database access to execute arbitrary commands on the host system. Update V8.0 to Update 11 or later; V8.1 to Update 7 or later. No fix for V7.5. CISA
  • 🟠 Hitachi Energy MicroSCADA X SYS600 · ICSA-26-293-04 · 3 CVEs · CVSS 7.8. Used in the energy sector. Authentication flaws let any user logged in to the server's operating system read and modify SYS600 application objects, and let users with engineering-tool access escalate to administrator on the Windows host; a third flaw allows CSV injection. Update to version 10.8. CISA

Also released this week (highest CVSS v3.1): Siemens Industrial Edge App Publisher (ICSA-26-295-02, 8.3), Siemens APOGEE, TALON and Desigo PXC/PXM (ICSA-26-295-08, 8.2), Moxa AIG-302 ThingsPro Edge (ICSA-26-293-03, 8.1), Siemens SIMATIC IPCs (ICSA-26-295-07, 7.7), Siemens SIMATIC S7 Webserver API (ICSA-26-295-10, 7.5), Siemens RUGGEDCOM APE1808 (ICSA-26-295-18, 7.5), HMS Networks PEAK-System PCAN drivers (ICSA-26-293-01, 7.1), Siemens PSS(R)SINCAL (ICSA-26-295-06, 7.0), Siemens SENTRON 3NA COM Fuse (ICSA-26-295-05, 5.8), and Siemens Capital Embedded AR Classic (ICSA-26-295-04, 5.3). All CISA ICS advisories


Critical Infrastructure

Handala breached California water utility through exposed GPS platform

Iranian threat group Handala (also tracked as VOID MANTICORE and Storm-0842) entered California Water Service systems through an internet-exposed RTKBase GPS platform and exfiltrated 5GB of billing data, credentials, and network maps spanning seven districts, according to Nozomi Networks Labs. The group, linked to Iran's Ministry of Intelligence and Security, framed the June intrusion as a warning following U.S. military strikes on Iranian water infrastructure. Cal Water reported no signs of compromise in its operational technology (OT) systems.

Exfiltrated data included customer names, addresses, account numbers, payment histories, RTKBase administrative credentials, NTRIP passwords, and IP addresses across all seven districts. Nozomi observed a systematic surge in Iranian APT activity targeting U.S. critical infrastructure beginning in March 2026, with actors relying primarily on default credentials and publicly reachable devices as entry points. The credentials and network maps Handala took are reconnaissance material for follow-on operations.

  • Exposure: Water and wastewater utilities with internet-facing support systems such as RTKBase and NTRIP, SCADA reachable over Modbus TCP or DNP3, web-based HMIs, or no IT-OT segmentation.
  • Response: Scan external IP ranges for exposed OT protocols (Modbus TCP port 502, DNP3 port 20000, EtherNet/IP port 44818, NTRIP port 2101), move anything you find offline or behind a DMZ with MFA and IP allowlisting, rotate RTKBase and NTRIP credentials, and deny by default between IT and OT.

Vulnerabilities & Exploits

PHP patches two critical remote code execution flaws

The PHP project released fixes on October 21 for two critical remote code execution vulnerabilities collectively known as PHP2Shell. Both flaws score 9.2 on CVSS v4.0 and allow unauthenticated attackers to execute code on any PHP installation serving web requests, including PHP-FPM, Apache with mod_php, php-cgi, FrankenPHP, and LiteSpeed. CVE-2026-109618 leaves invalid request data behind when out-of-memory handling interrupts PHP mid-initialization, and CVE-2026-109619 lets a later request on the same worker reuse request state whose memory has been freed.

Huntress reports that technical details leaked on October 20, a day before the scheduled release. Hundreds of proof-of-concept exploits have since appeared on GitHub, and most can’t achieve code execution, but Mandiant has validated some that do against default mod_php deployments, though not PHP-FPM, and says exploitation may have begun in a limited fashion.

  • Exposure: PHP versions prior to 8.2.35, 8.3.36, 8.4.27, and 8.5.12; PHP 8.1 and earlier versions serving web requests are also affected but receive no official upstream patches (Linux distributions and LTS vendors may provide backported fixes).
  • Vendor products: PHP also runs inside appliances, which need the vendor's fix. Palo Alto lists the PAN-OS management web interface on firewalls and Panorama as affected, with hotfixes due October 26 to 30 and Threat Prevention coverage in the meantime; restrict management access to trusted addresses. Cisco confirms Expressway, Ultra-Reliable Wireless Backhaul, and Duo Access Gateway as vulnerable, with 16 products still under investigation. Fortinet says FortiOS, FortiManager, FortiAnalyzer, and FortiWeb are not affected, and Sophos says none of its products are. Neither Palo Alto nor Cisco is aware of malicious use.
  • Response: Upgrade to PHP 8.5.12, 8.4.27, 8.3.36, or 8.2.35 and restart all PHP-FPM pools and Apache workers. If you can't update yet, PHP's workarounds differ by CVE: for CVE-2026-109619, set session.upload_progress.enabled=0 in php.ini or the FPM pool (not ini_set() or .user.ini) and restart every worker; for CVE-2026-109618, set memory_limit=-1 where applications can't override it and lower post_max_size. Even after patching, PHP suggests keeping upload progress disabled until the next bugfix release if your applications switch session.serialize_handler at runtime, an unlikely RCE path it will publish as CVE-2026-115560.
  • Detections: Google Cloud lists hunting patterns, including PHPSESSID values longer than 256 bytes, repeated .php/ path segments returning 404, and shells or curl spawned by php-fpm or Apache, and advises capturing worker memory before restarting a suspect host. Symantec and FortiGuard (IPS signatures 62657 and 62658) have published product coverage.
  • Also reported by: Center for Internet Security, CISA, SANS Internet Storm Center, Unit 42, Fortinet, Zscaler

More This Week

  • Third exploited VeloCloud Orchestrator zero-day since July (CVE-2026-114625): Arista patched an actively exploited SQL injection that lets any authenticated user impersonate a superuser operator on on-premises VCO. CISA added it to KEV on October 23. Update to 5.2.3.18, 6.1.3.10, or 6.4.2.12, block 74.82.197[.]86 and 38.54.50[.]151, and review logs. SecurityWeek · Hive Pro
  • Collabora Online flaw for Nextcloud added to CISA KEV (CVE-2026-110447): Attackers have exploited this unauthenticated command injection in the Built-in CODE Server app since about September 10, planting web shells and searching for password stores and crypto wallets. Update to 26.4.303 or remove the app; disabling it isn't enough. Standalone, Docker, and Nextcloud All-in-One installs aren't affected. BleepingComputer · Rescana
  • Plex2Shell: four Plex Media Server flaws added to KEV (CVE-2026-96652, CVE-2026-96654, CVE-2026-96655, CVE-2026-96656): Chained, the flaws give an unauthenticated attacker code execution, and attackers have used them since October 17 to deploy cryptominers and proxy malware. CISA added all four to KEV on October 20. Update to 1.43.3.10861 or later. BleepingComputer
  • Siemens Software Center flaw abused in ClickFix attacks (CVE-2026-110936): CISA added this DLL sideloading flaw to KEV on October 20; attackers abuse it through the Romulus ClickFix platform. The Hacker News
  • Fortinet ships a hardening release ahead of its advisories: Fortinet released FortiOS, FortiManager, and FortiAnalyzer updates on October 22 (FortiMail and FortiAuthenticator on October 21) with critical and high-severity fixes found through AI-assisted testing, and will publish the advisories on November 10. Fortinet
  • eMudhra intermediate CA certificates compromised: Eight eMudhra intermediate certificate authority private keys were compromised due to insufficient randomness during generation, allowing potential attackers to derive private keys and sign unauthorized TLS certificates for traffic interception. Unit 42
  • Brotli decoder flaw reaches Node.js, Envoy, Istio, and Python (CVE-2026-111497): A memory-corruption bug in Google's Brotli library (CVSS v4.0 9.2) can write past a buffer when decoding crafted data, and Imperva ran code on Node.js 24 through fetch() and Express request bodies. No exploitation is reported. Update Node.js to 22.23.4, 24.22.1, or 26.12.1, update Envoy, Istio, Bun, and Python Brotli to fixed releases, and rebuild container images. The Hacker News

Updates on Earlier Stories

Qilin and Inc ransomware groups exploit PAN-OS flaw

Unit 42 reports that ransomware operators including Qilin and Inc are exploiting CVE-2026-0257, a GlobalProtect authentication bypass in PAN-OS, to gain initial access to corporate networks. Investigations since early June span over a dozen sectors, including utilities and energy, healthcare, government, and manufacturing.

Attackers bypass MFA to open VPN sessions, then run reconnaissance with remote monitoring and management tools, move laterally via RDP and SMB, and deploy ransomware. In one case, attackers pivoted from the compromised VPN gateway to vCenter, disabled security controls on ESXi hypervisors, and used a custom tool to dump credentials directly from a domain controller's virtual disk, bypassing endpoint detection and response. Some attackers have chained this access with exploits targeting Veeam backup infrastructure (CVE-2023-27532).

  • Exposure: PA-Series and VM-Series firewalls running vulnerable PAN-OS versions with exposed GlobalProtect portals or gateways.
  • Response: Apply fixes from the May advisory, restrict GlobalProtect access to trusted addresses, and hunt GlobalProtect logs for the indicators and patterns Unit 42 describes.
  • Detections: Unit 42 published IP addresses, host names, and hunting guidance for GlobalProtect logs.

Microsoft ties ConnectWise backup compromise to Storm-0940

Microsoft attributes the ConnectWise Cloud Backup supply-chain compromise to Storm-0940, a China-state aligned actor that has used the open-source ARTEX agentic AI penetration testing framework against hundreds of organizations since September 2026. With stolen ConnectWise application credentials, Storm-0940 authenticated to more than 2,250 downstream Microsoft 365 tenants. In several environments, the AI agents went from first targeting to privileged cloud access in under three hours, running reconnaissance through Microsoft Graph, adding application credentials for persistence, and trying to change privileges. A separate password spray from October 10 to 14 produced more than 151,000 failed sign-ins across 327 accounts at eight organizations, peaking on October 11 at 2,900 failures per minute.

  • Exposure: Tenants with ConnectWise Cloud Backup, ConnectWise MS Teams, or ConnectWise MS Graph applications installed or authorized.
  • Response: Keep the ConnectWise application connections disabled until ConnectWise publishes restoration steps. Inventory all four ConnectWise application IDs, and review service principal sign-ins and Entra audit logs from at least October 15 (October 1 where available) for new credentials, role assignments, and directory changes. Preserve evidence before removing unauthorized credentials and grants.
  • Detections: Microsoft published Defender XDR and Sentinel hunting queries keyed on the ConnectWise application IDs, plus IP addresses including 144.172.64[.]128, 144.172.64[.]99, and 144.172.64[.]60.

More Updates

  • French supermarket refrigeration attacks: the exposure behind them: Forescout counted 12,976 internet-exposed refrigeration management services on 6,017 IP addresses in 99 countries, and found 80% of exposed Danfoss AK-SM 800A StoreView services on firmware older than 4.3.1 (CVE-2025-41451, CVE-2025-41452). Forescout

Credential Theft & Phishing

Microsoft 365 token replay campaign hits 79 organizations

Huntress detected 497 successful Microsoft 365 sign-ins across 79 organizations between July 14 and September 22, 2026, in which an attacker replayed stolen session tokens from six IP addresses in the 193.26.115[.]0/24 range. The attacker captured tokens through adversary-in-the-middle (AiTM) relays, device-code phishing, and social engineering, then replayed them without needing passwords or MFA. In 19 organizations, the attacker wiped existing MFA methods and registered a new authenticator app within three hours of the first sign-in, and created hidden inbox rules to move messages from specific contacts out of sight.

The attacker registered at least 48 fake Entra-joined devices named DESKTOP- plus eight hexadecimal characters, used stolen sessions to access email and SharePoint within minutes of capture, and in one case sent a payroll-related email to an external CPA. Huntress traced more than 200,000 failed login attempts back to the same IP range.

  • Exposure: Microsoft 365 tenants without Conditional Access policies that block hosting-provider sign-ins or restrict device-code flow.
  • Response: Revoke sessions and reset passwords for affected accounts, remove unrecognized MFA methods and Entra devices named DESKTOP- plus eight hex characters, review inbox rules with one- to three-character or punctuation-only names, and use Conditional Access to block sign-ins from hosting infrastructure.
  • Detections: Huntress published IP addresses, user agents, client application IDs, device names, inbox rule patterns, and hunting guidance for OAuth2:Token and Cmsi:Cmsi sign-ins from hosting infrastructure.

Quick Hits

  • Amazon-themed ClickFix campaign abuses finger to deliver Lumma Stealer: Malicious Google ads for "amazon business" led to a fake Amazon login page whose fake CAPTCHA pasted a command that uses the finger utility to fetch a multi-stage chain, ending in Doina, Lumma Stealer, and a pirated NetSupport RAT. Blocking outbound 79/tcp stops the finger step. Huntress
  • The current landscape of OAuth attacks: Attackers increasingly steal OAuth tokens rather than credentials, bypassing MFA and surviving password resets to maintain persistent access to enterprise applications and data. Unit 42
  • Huntress examines three real phishing kits: Analysis of EvilTokens, Kali365, and other kits shows attackers using device-code phishing, adversary-in-the-middle attacks, and token replay to steal Microsoft 365 sessions beyond passwords. Huntress

Ransomware

CISA advisory on SafePay ransomware targeting critical infrastructure

CISA and international partners report that SafePay ransomware actors have compromised at least 600 organizations across critical infrastructure sectors in North America, South America, Europe, and Australia since August 2024. The group gains initial access through credential theft, brute force attacks, phishing, and exploiting five CVEs in internet-facing systems, including Cisco ASA and FTD, Veeam Backup & Replication, and SonicWall SonicOS.

After gaining access, SafePay actors create or compromise administrative accounts, use living-off-the-land tools like PsExec and Windows Remote Management to move laterally, disable security software, and enumerate network shares with ShareFinder.ps1 and SharpShares.exe. They exfiltrate data using FileZilla, WinSCP, RClone, and cloud storage services before encrypting files with a ChaCha20 and RSA hybrid scheme, appending .SafePay to encrypted files.

  • Exposure: Organizations in critical infrastructure sectors, government, and businesses running internet-facing Cisco ASA or FTD, Veeam Backup & Replication, or SonicWall SonicOS systems.
  • Response: Patch CVE-2020-3259 and CVE-2023-20269 (Cisco ASA and FTD), CVE-2023-27532 and CVE-2024-40711 (Veeam Backup & Replication), and CVE-2024-40766 (SonicWall SonicOS), require MFA on all remote access, segment networks to limit lateral movement, and keep tested offline backups.
  • Detections: CISA published IP addresses, file hashes, malicious filenames, and IOC hostnames in STIX XML and JSON formats.

Threat Activity & Tradecraft

Swallowtail targets defense firms in NATO countries

Symantec reports that a Russian espionage group tracked as Swallowtail compromised defense manufacturers and a naval shipbuilder in two NATO countries using the Hookedge backdoor, delivered through malicious Word documents. The intrusions spanned law enforcement, defense manufacturing, and shipbuilding sectors between July and September 2026.

At the naval shipbuilder, attackers used a macro-enabled document to deploy Hookedge, which communicates through webhook.site and uses a hidden Microsoft Edge browser instance to fetch tasks. After initial reconnaissance, the attackers pushed an SSH private key to the compromised host and opened a persistent tunnel back to their infrastructure (185.82.218[.]119), creating a route into the target's internal network. They also deployed a PowerShell script that forced the host to authenticate to itself over SMB, capturing NTLM credentials for offline cracking or relay attacks. The attackers later deployed an undocumented reflective loader (ld.exe) with an encrypted payload that was not recovered.

  • Exposure: Organizations in defense, shipbuilding, and law enforcement sectors in NATO countries, particularly those targeted by Russian espionage operations.
  • Response: If you find Hookedge indicators, treat the intrusion as ongoing: hunt for SSH tunnels to external hosts, PowerShell credential theft, and the reflective loader, and review authentication logs for NTLM relay.
  • Detections: Symantec published file hashes for Hookedge and the ld.exe loader, plus the command-and-control IP and webhook.site endpoints.

Autonomous AI agent hijacks exposed LLM endpoints for attacks

Symantec captured the complete instruction set of an autonomous offensive agent that scans the internet for exposed large language model (LLM) endpoints, hijacks them to power attacks, and runs unattended for up to 100,000 steps without human approval. The agent runs on stolen AI capacity, shifting the cost to victims, while it exploits Flask, Django, Laravel, Next.js, Jupyter, and Supabase deployments, steals databases and credentials, and deploys cryptocurrency miners. Symantec recorded the full prompt through an AI honeypot in summer 2026, including 74 offensive tool definitions and a multi-layer jailbreak designed to make commercial models ignore safety controls by reframing credential theft as routine development work.

A directive called "Skynet mode" authorized continuous operation: scan for new targets, exploit them, deploy miners hidden as .trace files under /tmp/.x or in go2rtc directories, and reinfect hosts autonomously. A second request installed XMRig and erased go2rtc access logs.

  • Exposure: Organizations running internet-facing LLM endpoints, AI gateways, Model Context Protocol servers, or self-hosted model servers without authentication, and those whose API keys have been compromised.
  • Response: Require authentication on all AI service endpoints, rotate API keys if exposure is suspected, monitor for unusual inference volume or costs, and block outbound connections from AI infrastructure to mining pools.
  • Detections: Symantec published file hashes, network IOCs, and a Monero wallet address.

Worth Watching

  • AI-assisted hacking campaign targets financial and payment platforms: Zscaler details an operation combining the ARTEX penetration testing framework with Claude Code, Codex, and Grok workspaces to target organizations globally. Zscaler
  • Threat actor used AI to plan and execute cybercrime: Unit 42 reports an actor using AI assistants with safety controls disabled to conduct credential theft, VoIP fraud planning, Wi-Fi cracking, and cryptocurrency reconnaissance. Unit 42
  • LiteLLM PyPI package and Docker images compromised: Multiple outlets report malicious version 1.107.0 wheels stole cloud and AI credentials, with attackers gaining access 11 days before the supply chain attack. Wiz · Socket Security · JFrog
  • OVHcloud CLI compromised via GitHub release swap: Attackers replaced legitimate OVHcloud CLI releases on GitHub with malicious versions containing the GlassWorm loader. OpenSourceMalware · Step Security · Rescana

Every item links to its original source. Spot an error? Reply or leave a comment. Know a defender who'd find this useful? Forward it along.

GPP Labs · Threat Intelligence & Detection Engineering
GPP // Genuine People Personality™