Case · DiggingBeagle record
Check Point turned an incomplete DeepSeek-attributed sample into a controlled Android proof of concept that used legitimate browser folder permissions to encrypt selected images without a native payload.
Defensive research and proof of concept. Check Point says it had not observed the exact browser-native technique adopted in the wild.
The original sample was noisy. It claimed capabilities that a normal web page cannot perform and mixed browser-limited stubs with malware-style features. One part was different: it reached for the File System Access API, a legitimate browser capability that can read and modify files after a user grants folder access.
Check Point used that idea to build a controlled proof of concept on Android Chrome. A fake image-processing workflow gave the victim a plausible reason to approve folder access. After the permission was granted, the page could encrypt pictures inside the selected directory without installing an APK, exploiting the browser or obtaining root access.
The research is useful because it shows a model connecting an unrealistic malicious objective to a real platform primitive. It is not evidence of a deployed ransomware campaign. Check Point explicitly says the original sample was incomplete and that it had not observed the exact browser-native pattern in the wild.
A page gives the user a plausible reason to process local images.
Boundary: social engineering / browser permission
The browser exposes a legitimate File System Access permission prompt.
Boundary: web page / local files
The page gains read/write access to the selected directory after approval.
Boundary: permission / file authority
The controlled proof of concept modifies files through the granted browser API.
Boundary: browser capability / local damage
The report documents the incomplete sample and controlled Android proof of concept.
Locator: In-Browser Ransomware on Android
The controlled proof of concept used the File System Access API after user approval to encrypt pictures in a selected directory.
Locator: Key takeaways and Noisy Sample section
Check Point says the original sample was incomplete but identified a practical browser-native file-access path.
Locator: Introduction and telemetry dataset
The research reports nearly 3,000 files analyzed and 1,383 classified as malicious or dangerous.
Scope: Absence of evidence in the Check Point dataset is not proof that no independent use exists.
Locator: Conclusion
The researchers explicitly say they had not observed the exact technique in widespread real-world campaigns.
diagram
Victim expects local image processing
Browser asks for directory access
Page can read and modify selected files
Controlled PoC overwrites selected images
Browser permission prompts can be security boundaries when web applications gain durable file-write access. AI lowers the effort required to discover and combine those primitives, but a controlled PoC should remain clearly separated from incident evidence.
DiggingBeagle. “An AI-generated malware idea exposed a workable browser-only ransomware path.” First seen Jul 1, 2026. https://diggingbeagle.com/cases/an-ai-generated-malware-idea-exposed-a-workable-browser-only-ransomware-path/
Citation guidanceWhy this archive exists
DiggingBeagle is a non profit research project documenting AI security incidents, agent failures, vulnerabilities and AI-assisted operations. A case keeps its claims beside the sources that support, contest or limit them. Later updates stay visible, so a reader can see when the account changed.
We publish case reconstructions, dated reporting and analysis across records. Each has a different evidentiary role. About the project and our methodology explain how the work is reviewed.