Week of October 5, 2026

Cyderes announced a partnership with Palo Alto Networks to integrate its managed detection and response (MDR) services with Cortex XDR, Cortex XSIAM, and Idira. The collaboration aims to modernize AI-driven security operations for enterprises across the US, Canada, and U.K./EMEA.

A vulnerability identified in CellHubs ExifTool 5.0.1-gms allows OS command injection via the CSV export functionality. This could enable an attacker to execute arbitrary commands on the system, potentially leading to full compromise.

Blind Eagle, a threat actor active in Latin America, is deploying Remote Access Trojans (RATs) to compromise systems. The campaign highlights evolving TTPs and an expanding e-crime arsenal, posing significant risk to organizations in the region.

Cisco Talos analyzed ROZESHELL, a malware family that includes AI-evasion comments, AMSI bypass, runtime compilation via csc.exe, and a Rozena shellcode loader. The anti-analysis text is not necessarily a marker of unsophistication; it can accompany complex behavior. PLOTSAFE evolves from fixed sentences to generated variations.

The FakeGit campaign has reactivated, now using over 17,000 fake GitHub repositories to distribute SmartLoader malware. This campaign previously pushed StealC infostealer and now targets developers and users downloading from GitHub.

A pre-authentication root remote code execution vulnerability, dubbed AnyPwn, affects AnyDesk for Linux version 8.0.2. The flaw is a heap overflow that can be exploited without authentication, allowing an attacker to execute arbitrary code as root. Users should update to a patched version immediately.

UAC-0099 (a threat actor) is targeting Ukrainian organizations with a new malware variant called MATCHBOIL. The malware uses two-minute C2 polling intervals and sandbox evasion techniques to avoid detection. This incident highlights evolving evasion tactics against endpoint security tools.

Kaspersky announced AI Protect, a new security solution that scans large language models (LLMs), AI agents, and AI infrastructure for threats such as model poisoning and malicious files. The product aims to secure AI deployments by detecting and mitigating risks before they become active threats.

IBM Guardium Data Protection 12.2 contains a hard-coded recovery key and weak cryptographic protection in the pkcrypto passkey component. A local attacker can exploit this to decrypt sensitive data. This vulnerability undermines the confidentiality of data protected by the product.

Warden Stealer is a Rust-based infostealer that has rapidly gained popularity, targeting cryptocurrency wallets, browser data, and AI agent credentials. It is distributed via phishing and malvertising, and its technical capabilities include stealing from over 100 crypto wallets and exfiltrating data via Telegram. The stealer's focus on AI agent data represents an emerging threat vector.

CrowdStrike reported that a Chinese-speaking threat actor used the agentic pentesting tool ARTEX, integrated with Claude, to automate reconnaissance and exploitation against South Korean financial institutions. The tool enabled the attacker to efficiently breach systems and exfiltrate data.

Splunk patched a critical vulnerability (CVE-2026-76268) that allowed unauthenticated remote command execution via the Patroni REST API. The flaw could permit attackers to execute arbitrary OS commands on affected systems.

Microsoft tracked a campaign using counterfeit software installers from spoofed vendor pages to deliver a self-protecting, persistent implant. The attack chain includes initial access, delivery, execution, persistence, privilege escalation, defense evasion, and command and control.

Attackers used AI coding tools to develop and execute attacks against South Korean online banking infrastructure. CrowdStrike observed interactive sessions where the operator fed target environments into AI interfaces, weaponizing a local testing framework against live systems.

Stellar Cyber announced version 7.0 of its Open XDR platform, adding queue-level AI triage, Microsoft Defender for Endpoint actions, Microsoft 365 and Entra ID detections, and remote uninstall for connected server sensors. The release targets MSSPs with features to streamline operations and expand detection coverage.

Airlock Digital released version 4.1 of its free Allowlist Auditor utility, which tests application allowlists for misconfigurations, weaknesses, and bypass techniques. The tool helps organizations assess the effectiveness of their application control policies.

A malware botnet named PoeLLM has compromised more than 3,400 servers. The malware hides its command-and-control infrastructure coordinates within a poem. This technique allows the botnet to evade detection and maintain persistence.

Elastic published 14 security advisories addressing vulnerabilities in Elasticsearch, Kibana, and Elastic Agent/Endpoint. One flaw enables cross-tenant data interception, potentially exposing sensitive data across organizational boundaries. The patches mitigate risks including unauthorized access and data leakage.

A vulnerability in Veeam Backup & Replication allows a low-privileged user with Backup Viewer role to modify or delete the Enterprise Manager master key and read or overwrite stored antivirus update credentials on the backup server. This can lead to remote code execution. The flaw is rated medium severity with a CVSS score of 6.1.

An organization experienced an Akira ransomware attack in September. Investigation revealed the threat actor's tactics and tools, including the use of a specific malware. The incident highlights the evolving techniques of the Akira ransomware group.

On the first day of Pwn2Own Ireland 2024, ethical hackers disclosed 32 zero-day vulnerabilities across various products, including endpoint security solutions. The event highlights weaknesses in widely deployed software and provides vendors with details to develop patches.

Kevin Mandia, founder of Mandiant, announced his new startup Armadin and stated that AI has discovered over 90 zero-day vulnerabilities in Fortune 500 companies. He warns that AI has crossed a threshold where it can autonomously find vulnerabilities, and defense must become autonomous to keep pace.

A critical vulnerability in WatchGuard endpoint security products allows a local authenticated attacker to bypass driver authentication and access sensitive kernel and process memory. The flaw resides in the endpoint security agent's driver, potentially leading to privilege escalation or information disclosure. WatchGuard has released a patch to address the issue.

Microsoft observed ClickFix attacks using browser cache smuggling to execute cached VBScript and launch a credential-targeting malware chain. The technique bypasses Windows Run limits by loading malicious scripts from the browser cache instead of directly executing downloaded files.

A 2024 ransomware incident demonstrated attackers pivoting undetected from one asset to another despite EDR deployment. The attackers hid malicious code in a webcam's firmware or storage, evading endpoint detection. This highlights a gap in EDR coverage for peripheral devices.

A vulnerability in Rejetto HTTP File Server (HFS) was discovered by Horizon3.ai's Mythos AI model, which used mathematical reasoning to reverse the PRNG and reconstruct the secret session-cookie signing key. The flaw was patched in HFS version 3.2.1 released on July 13, but exploitation has been observed in the wild.

Trend Micro announced TrendAI Autonomous Vulnerability Discovery, a new service that combines AI-powered vulnerability discovery with validation from a large vendor-agnostic bug bounty program. The service aims to identify exploitable attack paths and provide next-generation threat remediation at machine speed.

Attackers are configuring Microsoft Defender exclusions to hide malicious directories and file types while keeping antivirus visibly enabled. This technique requires administrative access and is used post-compromise to evade detection.

Microsoft warns that attackers are using fake CAPTCHA verification pages to trick users into executing malicious commands on Windows systems. The technique, known as ClickFix, leverages social engineering to bypass security controls and deliver malware. This matters because it demonstrates a shift toward human-operated deception that can evade traditional endpoint detection.