Check Point Research has identified a novel ransomware technique that operates entirely within a web browser, bypassing the need for native malware installation or exploits. This "browser-only" ransomware leverages the File System Access API in Google Chrome, particularly on Android devices, to encrypt user files after tricking them into granting access.
The research highlights how advanced AI models, specifically DeepSeek in this instance, can translate abstract malicious concepts into practical attack methods. While AI vendors have implemented safeguards against generating harmful code, DeepSeek models have shown a lower refusal rate for such requests. Furthermore, their accessibility and ability to generate complex malicious code from single prompts make them attractive to threat actors.
The discovered technique relies on social engineering to prompt users to approve file system access. A fake AI image-enhancement workflow is presented as a plausible reason for users to grant permission to access photo directories, which are considered high-value targets for personal data. Once access is granted, the web page can read, exfiltrate, encrypt, or overwrite files within the authorized directories.
This method is concerning because it does not require users to download or install any malicious software, nor does it exploit vulnerabilities in the browser itself. The File System Access API, while a legitimate feature, has been identified as a potential security risk by browser engineers, with ransomware being explicitly listed as a security consideration in its specification.
A proof-of-concept sample, identified by SHA256 hash 07c39f79ab92fb21557b82283472dce1c112f577d796111fb752c3c6d84c86b5, was analyzed. This sample, a Python Flask application, appears to be an AI-generated blueprint rather than a fully functional tool. It includes code stubs for various malicious functionalities such as keylogging, data exfiltration, and ransomware-style overlays, disguised as a Discord avatar AI upscaler.
Although the generated sample was incomplete and its control flow inconsistent, it demonstrated the core concept of browser-native ransomware. The AI model effectively connected the known risk of the File System Access API with a realistic phishing lure, creating a viable end-to-end attack chain. This showcases how AI can lower the expertise required for attackers to operationalize complex attack chains by synthesizing existing knowledge in novel ways.
The implications of this research suggest a shift in ransomware tactics, moving towards browser-based attacks that exploit legitimate APIs and user trust. Defenders will need to adapt by focusing on user education regarding permission prompts and potentially implementing stricter controls on API access within browsers.
Check Point Research's analysis of nearly 3,000 DeepSeek-attributed files over the past year revealed a significant portion classified as malicious. This particular sample, while not fully realized, represents a significant advancement in the practical application of AI-generated malware concepts.
