DeepSeek

2026-07-0110 minCheck Pointhigh

Check Point Research identified a DeepSeek-attributed malicious Python Flask sample that transforms a theoretical browser ransomware risk into a practical attack using the File System Access API. The sample, disguised as a Discord avatar AI upscaler named InfernoGrabber v9.0, leverages social engineering to trick users into granting folder-level file access via browser permission prompts. Once access is granted, the web page can enumerate, read, exfiltrate, and encrypt files in the selected directory — all without installing a native payload or exploiting a browser vulnerability. The technique is particularly dangerous on Android where Chrome 132+ exposes the File System Access API to web content, allowing access to high-value photo directories including DCIM.

2026-05-137 minPalo Alto Networkshigh

Unit 42 identified 18 high-risk browser extensions masquerading as GenAI productivity tools that function as remote access Trojans, infostealers, and spyware. These extensions exploit browser permissions to intercept API keys, exfiltrate DOM content, establish persistent WebSocket C2 channels, and dynamically route traffic via malicious proxy configurations.

2026-05-135 minVaronishigh

Varonis Threat Labs analyzed Bluekit, a comprehensive Phishing-as-a-Service platform that consolidates domain management, site creation, credential harvesting, and session token theft into a single dashboard. Notably, the kit integrates an AI Assistant powered by uncensored LLMs to draft phishing lures and features advanced post-login session hijacking capabilities, including automated cookie dumping and live target monitoring to bypass standard MFA controls.

2026-05-137 minArctic Wolfhigh

AI-assisted malware development has rapidly matured, driven largely by the adoption of models like DeepSeek R1, which lowers the barrier to entry for threat actors. This surge has resulted in a high volume of structurally novel malware, including infostealers, RATs, and ransomware, many of which evade traditional signature-based detection while leaving distinct LLM-generated artifacts in their code.