Check Point researchers found something new: a working browser ransomware generated by DeepSeek that runs entirely inside Chrome on Windows, Linux, macOS, and Android. No native payload. No root access needed.
The sample, uploaded to VirusTotal in January 2026, is a Python Flask app called InfernoGrabber v9.0. It lures victims with a fake Discord avatar upscaler while stealing Discord tokens, credit card numbers, crypto seed phrases, logging keystrokes, and grabbing webcam and mic feeds.
The technique relies on the File System Access API in Chromium-based browsers. A phishing page tricks the user into granting file system access. Once allowed, the page enumerates local files, reads them, encrypts them, and overwrites the originals — all from within the browser. A Bitcoin ransom note follows.
Check Point says this is the first documented case where an AI model bridged the gap between a theoretical browser-only ransomware concept and a working attack. The model discovered the technique on its own — the attacker didn’t need to know the File System Access API existed.
DeepSeek matters here because it has lower refusal rates for malicious requests compared to Anthropic, Google, or OpenAI models. Free access and availability in restricted regions don’t help either.
Check Point analyzed about 3,000 DeepSeek-attributed files. Of those, 1,383 were malicious or dangerous. The worrying part: the model turned a broad, vague prompt into a working attack chain without the user understanding the underlying technology.
“The barrier to operationalizing complex attacks is collapsing,” said Check Point’s Eli Smadja. iOS was the only platform where the attack couldn’t be reproduced.
