ThreatsDay: Adobe RCE, DNS Wallet Theft, and NVIDIA Container Escapes
A weekly breakdown of critical vulnerabilities, evolving malware, and emerging cyber threats.
1GlobalProtect VPN Exploitation
Qilin and Settra ransomware groups are actively exploiting CVE-2026-0257, an authentication bypass in PAN-OS GlobalProtect. Since the vulnerability allows access through an organization's own VPN, malicious activity often mimics legitimate remote work. Security teams must patch immediately, terminate all active sessions, and investigate suspicious connections, as patching alone may not remove existing unauthorized access.
2Cryptocurrency Wallet Drainer Kits
The Noir wallet-drainer kit utilizes DNS TXT records on Cloudflare Pages to locate its active backend pool. By requesting spending permissions rather than direct transfers, the malware appears less suspicious to victims. It is suspected that AI assistance was used to develop and maintain this kit, which operates without needing passwords or private keys.
3Psychological Warfare in Malware
The KGB RAT, marketed since late 2025, distinguishes itself with built-in psychological harassment features. Beyond standard credential theft and ransomware capabilities, it includes screamer functions, jump scares, and forced blue screens of death. This strategy weaponizes victim anxiety, signaling a shift toward threat actors who prioritize psychological impact alongside traditional data exfiltration and persistent control.
4BotHelper RAT Surveillance
BotHelper RAT is a multi-stage .NET malware that performs host fingerprinting before retrieving its second-stage payload from easyllms.xyz. It registers a scheduled task to relaunch every 30 minutes and patches the Antimalware Scan Interface to evade detection. The malware maintains persistence by polling PHP endpoints for tasking while actively monitoring the infected host's live screen.
5Android Scam Detection Upgrades
Google is deploying multi-layered protections for Android, including Gemini Nano-powered link safety checks and real-time bank call verification. Since June 2026, the system has blocked over 3 million spoofed calls from Banco do Brasil and Itaú Unibanco. These features automatically hide sensitive information during screen-sharing and prevent apps from silently enabling call forwarding in the background.
6AI Model Distillation Defense
Anthropic has implemented new measures, including preserved thinking and two-stage classifiers, to protect its Claude models. By charging for requests blocked by safeguards, the company has reduced data harvesting through distillation attacks by 90% to 99%. These updates apply to the Fable 5.1 and Opus 5.5 models, significantly hardening them against unauthorized data extraction attempts.
7WordPress Self-Healing Malware
A sophisticated WordPress malware uses must-use plugins and EtherHiding techniques to survive removal attempts. By hiding command-and-control server locations behind Ethereum smart contracts, the malware ensures its infrastructure remains resilient to takedowns. It grants attackers administrator access, hides rogue accounts, and facilitates data exfiltration while spreading to other WordPress installations on the same file system.
8XWorm Surveillance Evolution
The XWorm malware has transitioned into a comprehensive surveillance platform, expanding its plugin count from 35 to 46. Recent versions, including V7.4, enable live microphone streaming, GPS location tracking, and webcam capture. Its HBrowser plugin allows operators to control authenticated browser sessions directly, effectively bypassing multi-factor authentication without needing to steal the underlying user credentials.
9Chrome Memory Safety Bounties
Google has quadrupled base rewards for memory safety vulnerabilities in Chrome to $2,000. While standard payouts for some bugs were lowered in April, the focus on memory safety reflects a strategic shift in bug bounty priorities. Researchers can now earn up to $20,000 for identifying critical vulnerabilities within the browser's GPU, network, or core processes.
10AI Assistant Credential Theft
Infostealers like Aoki and Warden Stealer are increasingly targeting configuration files and session tokens for AI coding assistants. By stealing API keys for tools like Cursor, Claude, and Gemini, attackers gain access to repositories and CI/CD pipelines. These credentials are highly lucrative, allowing threat actors to run unauthorized workloads or resell access on the illicit LLM market.
11Harbor and AEM Vulnerabilities
A critical SSRF vulnerability in Harbor allows registered users to steal cloud IAM credentials by pointing webhooks at metadata endpoints. Simultaneously, Adobe Experience Manager (AEM) cloud versions faced a four-part RCE chain involving unsafe XML parsing and Groovy execution. Both issues have been patched, highlighting the risks associated with container registries and complex enterprise content management systems.
12NVIDIA GPU Driver Exploitation
Researcher Ziyi Guo disclosed a critical Linux vulnerability in NVIDIA's GPU Display Driver, known as Bad UVM. By exploiting improper memory permission handling, an unprivileged user can achieve code execution or container escape. This flaw demonstrates the severe security implications of driver-level vulnerabilities, which can bypass standard isolation boundaries and grant attackers deep control over the host system.