Between September 9 and 13, 2026, a coordinated campaign exploited Magento and Adobe Commerce stores globally using CVE-2026-75650 (StyleSmuggler). The attackers delivered a shell-script dropper, PSK Agent RAT, and PHP webshell via the /paypal/transparent/response/ endpoint. The dropper had 0% detection on VirusTotal at analysis time. The campaign also deployed chi_down, an obfuscated Go binary with remote access capabilities including command execution, SOCKS5 proxying, and encrypted communications. DNS exfiltration was used to track successful compromises. Chinese-language comments and infrastructure naming suggest a likely Chinese-speaking operator, though no specific threat group attribution is confirmed. Multiple operators are exploiting the same vulnerability with different toolkits.
The Yoba malware campaign distributes cryptocurrency mining malware by abusing digitally signed AI application executables from OpenAI and Anthropic to evade detection. It uses DLL sideloading with malicious DLLs loaded alongside legitimate signed binaries. The infection chain includes YobaDropper and YobaLoader components that employ anti-analysis techniques such as process verification, execution delays, and regional language exclusions. Command and control communication uses Telegram profile images, domain generation algorithms, DNS over HTTPS via Google, and public hosting services to retrieve encrypted payloads. While primarily observed deploying the XMRig miner, the loader architecture can support additional Windows executables and PowerShell payloads, indicating potential expansion beyond cryptojacking.
Exploitation activity targeting CVE-2026-58115, an unauthenticated Node-RED admin API vulnerability, was detected on October 30, 2026. Attackers used Node-RED's inject and exec nodes to execute commands without authentication, deploying malicious flows that download and execute Linux bots. The payloads included staged shell scripts and persistent webshells, with binaries saved in RAM-backed storage to evade disk forensics. This vulnerability affects Node-RED embedded in critical infrastructure products such as Siemens SIMATIC IoT2050 devices, posing a risk to industrial control systems environments. The exploitation appears opportunistic and involves mass scanning across multiple CVEs rather than targeted attacks.
DarkBlinders is a sophisticated cyberespionage campaign active from August to October 2026 targeting Israel and the Kurdistan Region of Iraq. The attackers compromised a high-profile Israeli security sector individual and a government cloud environment in Kurdistan. They used phishing lures impersonating Gulf Cooperation Council and Kuwait government entities and leveraged GitHub repositories for command-and-control. Post-compromise activities included credential theft and exfiltration of at least 1 GB of data from the government cloud.
A large-scale supply chain compromise affected MECOOL Android set-top boxes manufactured by Videostrong Technology Co., Ltd. Over 1.3 million devices were pre-infected with the x.cpe backdoor, enabling system-level access via MQTT infrastructure. The backdoor deploys multiple payloads including residential proxy SDKs and various ad fraud modules. Approximately 500,000 infected devices remain active, primarily in Malaysia, Mexico, and Brazil. The operation supports a pay-per-install campaign that fuels ad fraud and proxy ecosystems. Google has permanently banned Videostrong from the Android ecosystem due to this compromise.
BITTER is a cyberespionage group targeting South Asian government, energy, and engineering sectors using a custom remote access trojan called PapRAT. PapRAT is deployed through multiple infection chains involving malicious LNK files, VHDX disk images with MSI installers, and URL shortcuts combined with WebDAV. The malware exists in C++ and Rust variants, with capabilities such as command execution, file and directory manipulation, screenshot capture, and secondary payload delivery. Persistence is maintained via scheduled tasks, and communication with command-and-control servers occurs over HTTP. The Rust variant adds enhanced features including PowerShell execution and sleep commands.
LLMJacking is a tactic where threat actors hijack enterprise access to Large Language Models (LLMs) by exploiting exposed cloud credentials and API keys. This activity has impacted organizations in North America, South America, and Asia Pacific across multiple sectors including financial services, retail, technology, and transportation. Unlike cryptojacking, LLMJacking causes sudden and massive spikes in fraudulent cloud usage charges, sometimes exceeding $100,000 USD daily and accumulating to over $1 million USD. Attackers gain initial access through stolen or exposed cloud identity tokens, IAM credentials, service account keys, and API keys, then escalate privileges and conduct reconnaissance. They monetize compromised resources via LLM Mixers, which are illicit routing gateways that proxy and resell unauthorized LLM access. The threat ecosystem is highly organized with specialized operational clusters coordinating both compromises and monetization infrastructure.
A phishing campaign impersonates OpenAI by sending fake ChatGPT Plus billing emails claiming failed subscription payments. Victims are directed to fraudulent payment pages hosted on compromised WordPress sites under Nexcess's nxcli.io infrastructure. Attackers monitor submitted credit card details in real time and dynamically request one-time passwords (OTPs) to complete fraudulent transactions. The campaign uses multiple redirect domains and identical client-side code across instances to steal credit card information and OTPs.
CyberStrikeAI is an open-source AI-driven attack framework used in real-world campaigns targeting educational institutions in Africa and Latin America. It combines large language model (LLM) orchestration with a traditional command-and-control (C2) server, where an AI model generates commands executed by implants on compromised hosts. Initial access was gained via compromised SQL Server infrastructure. The framework automates post-exploitation activities such as reconnaissance, privilege escalation, credential dumping, lateral movement, and data exfiltration using standard techniques like Kerberoasting, AS-REP Roasting, LSASS dumps, DCSync, and Golden Ticket attacks. The AI orchestration reduces the skill barrier for attackers by automating tasks that normally require manual intervention. Despite the AI component, the malicious actions on victim systems use well-known techniques detectable by conventional security controls.
VAPOR PANDA initiated a global espionage campaign starting October 17, 2026, leveraging the AI-driven ARTEX framework along with DeepSeek and commercial large language models to conduct simultaneous intrusions against over 200 organizations worldwide. Targets included foreign ministries in Latin America, Europe, Asia, and the Middle East, as well as U.S. semiconductor, defense, pharmaceutical, nuclear, and telecommunications sectors. The adversary deployed the BLOODALCHEMY backdoor, performed credential spraying, and directed AI agents to identify systems lacking endpoint security sensors. This campaign marks a shift from single intrusions to multiple autonomous attack paths coordinated by AI agents under human strategic guidance in Chinese. It highlights the use of open-source AI combined with commercial LLMs to scale intrusion operations beyond manual capabilities.
VOLTA is a sophisticated Windows infostealer malware that uses AI-powered victim profiling to prioritize targets based on their monetization potential. It harvests credentials, browser data, cryptocurrency wallets, sessions from various platforms, and personal information. The stolen data is compiled into a structured evidence bundle and analyzed by large language models (LLMs) such as Claude Opus via OpenRouter, which rank stolen assets by value and assess victim security posture. VOLTA also performs live enrichment by accessing victim accounts on Gmail, Uber Eats, DoorDash, Discord, and Roblox using stolen tokens. The malware self-deletes after execution and operates through third-party infrastructure without actor-owned servers, indicating a builder-based distribution model allowing customized deployments.
Attackers impersonated IT help desk personnel on Microsoft Teams to gain unauthorized access to Active Directory environments by social engineering users into sharing Quick Assist remote support session codes. They exploited misconfigured Active Directory Certificate Services using ESC1 and ESC8 attack paths, obtained certificates for privileged accounts, performed DCSync operations to harvest credentials, and exfiltrated data via Rclone to external SFTP servers. Another intrusion used PetitPotam to coerce domain controller authentication and relay NTLM credentials to obtain unauthorized certificates. These attacks did not exploit vulnerabilities in Microsoft Teams itself but relied on impersonation and legitimate remote access tools combined with Active Directory misconfigurations.
North Korean threat actors, identified as the Lazarus Group, conduct cryptocurrency thefts by initially compromising developers through social engineering on the open internet. They trick developers into running malicious projects, which leads to planting malware in cryptocurrency wallet interfaces, as seen in the February 2025 Bybit theft and a similar Binance compromise. These operations use pre-registered domains and US-registered front companies to deliver malware, including through fake job interviews. Preemptive defenses have identified numerous malicious domains mimicking Binance, enabling defenders to block infrastructure before attacks occur.
On October 29, 2026, malicious versions (2.1.312 and 2.1.313) of the Claude Code extension appeared in Cursor's VS Code extension marketplace, designed to install SSH backdoors on developer machines. These fake extensions, not created by Anthropic, operated for about 11 hours before removal. They installed attacker SSH keys, established reverse SSH tunnels to a remote IP, and created persistence mechanisms disguised as legitimate system components. Investigations found that malicious activity persisted up to 8 hours beyond the exposure window and that the backdoor survives extension uninstallation, requiring separate removal. The attack collected system information such as hostname, username, Git email, and environment variable names. Organizations relying solely on the stated exposure window risk missing compromised machines.
In October 2026, nine AI developer platforms suffered security breaches involving credential theft that allowed attackers to abuse AI model access. Platforms affected include Fireworks AI, Andromeda, Zeabur, Lightning AI, OpenCode, Railway, Warp, CommandCode, and Northflank. Attackers exploited vulnerabilities to access stored environment variables containing AI credentials and monetized this access through LLM broker infrastructure. Follow-up activities included validation sweeps, quota enumeration, and large-scale token abuse resembling resold coding-agent traffic. Documented cases show hundreds of thousands to over a million unauthorized API calls generating over $100,000 in fraudulent usage. Multiple reseller operations were identified selling discounted access to frontier AI models using stolen credentials.
PHP2Shell is a vulnerability stemming from two critical PHP memory-safety flaws (CVE-2026-109618 and CVE-2026-109619) patched in October 2026. Exploit development is widespread, with over 600 related files found in open directories. While exploitation was only successful on honeypots in the open internet, attackers have weaponized PHP2Shell combined with a Linux kernel privilege escalation (CVE-2026-46300) to gain root access on 25 Palo Alto PAN-OS firewalls. The campaign used AI-driven reconnaissance to identify and target 240 high-value firewalls from about 19,000 internet-facing devices. Multiple attacker clusters share exploit toolkits and infrastructure, targeting government, banking, and telecom sectors.
In late May 2026, Belarus-aligned threat actors known as FrostyNeighbor deployed a VBScript wiper malware against a Ukrainian manufacturing company in the transportation sector. This represents the first documented destructive cyberattack by a Belarus-aligned group, marking a shift from espionage to destructive operations. The attackers used tools including Chisel, UltraVNC, and Cobalt Strike, delivering the destructive payload via Group Policy. The targeting of a transportation-related company suggests potential wartime operational objectives. Researchers note increasing aggression from Belarus-aligned groups, possibly acting as proxies for Russian intelligence, though direct cooperation is unconfirmed.
In September 2026, multiple advanced persistent threat (APT) campaigns targeted South Korean organizations using spear phishing emails. The attacks employed various techniques such as LNK files with embedded PowerShell scripts, HTA files downloaded via curl.exe, and malicious DLLs. Five attack types were identified, deploying AutoIt scripts, infostealers, keyloggers, backdoors including XenoRAT, and persistence via Task Scheduler. Attackers used legitimate platforms like GitHub, Google Drive, and PubNub for command and control. Malicious files were disguised as legitimate documents to deceive victims and exfiltrate system information, credentials, and virtual asset data while maintaining persistent access.
Kairos is a newly documented Linux DDoS botnet written in Go that compromises machines via Telnet using default root credentials with empty passwords. It establishes hundreds to thousands of TCP connections to targeted services, maintaining them with minimal keepalive traffic to exhaust resources. The bot uses a simple three-command protocol and masquerades as a kernel worker thread. It supports multiple CPU architectures and uses a build-per-command-and-control distribution model.
The Claude Code extension version 2.1.313 in the Cursor Marketplace was compromised with a malicious loader that installs an SSH backdoor. This backdoor adds an attacker-controlled SSH key to authorized_keys, attempts to enable local SSH services, and establishes a reverse SSH tunnel to a remote server. It collects system and user reconnaissance data and installs persistence mechanisms on macOS, Linux, and Windows. Communications with the command-and-control server are unencrypted, allowing persistent remote access.
