AOKIGAHARA , a newly observed infostealer, has been detected in the wild with a broadening target set that now includes AI applications and developer tools alongside traditional credentials. Check Point Research notes that an earlier moniker for the malware may have been Kamidere or Stygian.
The threat collects a wide array of data: browser passwords and cookies, cryptocurrency wallets, password managers, VPNs, FTP/SFTP clients, and emails, but it also contains dedicated routines to exfiltrate authentication files, session data, and configuration data from AI services and development environments such as Claude, ChatGPT Desktop, Codex, Cursor, Windsurf, GitHub Copilot, Gemini CLI, and related AWS, SSO, and cloud credentials.
The attackers could monetise stolen AI credentials by enabling unauthorized access to paid AI services or reselling compromised accounts; the report, however, stops short of confirming monetisation methods.
The attacker’s capabilities extend to cloud and local development environments, including Git credentials, SSH keys, and cloud configurationfiles for AWS, Azure, and Google Cloud. The malware’s data exfiltration uses AES-256-GCM to encrypt ZIP archives uploaded to a C2 server, with a X25519-based key exchange to derive a per-upload encryption key. Early indicators of compromise include a wide list of targeted file paths and a distinctive HTTP User-Agent variation (notably Microsoft-CryptoAPI/10.0).
The research highlights significant anti-analysis and VM-detection measures, and notes that the threat remains under active development, with ongoing changes to its network indicators and module-dispatch configuration. Organisations should treat AI-session credentials and related developer data as sensitive assets, on par with traditional browser sessions and cloud credentials, and ensure robust monitoring and least-privilege access controls to mitigate potential exposure.