← Lab /Security & privacy

Aokigahara stealer harvests AI tool credentials and session tokens

Security & privacy Measurement · Real-world · Check Point Research

The claim

The authors assert that the Aokigahara infostealer collects authentication data from AI applications and developer tools.

antoniost@checkpoint.com · Check Point Research Writeup

Published as: “Infostealer riding the AI Train”

So what

AI service credentials are now a high-value target for infostealers, expanding the threat landscape beyond traditional browser and wallet data.

The five-minute version

The writeup analyzes samples of the Aokigahara infostealer, detailing its many data-collection modules and its encrypted configuration. It highlights a dedicated "aisessions" handler that extracts credentials and session files from AI services such as Claude, ChatGPT Desktop, Codex, and GitHub Copilot, as well as cloud credentials. The authors confirm the presence of these handlers in real samples but note they have not observed how the stolen data is monetized. Evidence comes from static analysis of the binary and network traces of its C2 communication.

What they found 1

What undercuts it the paper’s own results

Stated limits 2

What’s new here 4

Artifacts & methods 3

Topics

infostealercredential theftAI application tokensX25519 key exchangeAES-GCM encryptionanti-analysis checksmodule maskC2 protocol
← Previous Malicious server can infer client label sets from FL gradient guidance (50-84% accuracy)