1-minute read
What a defender needs to know
Check Point Research identified Aokigahara, a Windows infostealer spotted in the wild that collects AI-assistant sessions, browser credentials and developer secrets. Its aisessions handler targets Codex .codex\auth.json, Claude .claude\credentials.json, Cursor storage, Gemini CLI configuration and AWS files under .aws. A separate ide handler targets Git credentials, .ssh material and cloud configuration. A targeted file may contain settings rather than a usable secret; finding that file alone is not a malware alert.
The malware packages each enabled handler’s data separately and uploads encrypted ZIP archives through /gate/chunk. Hunt for Check Point’s sample SHA-1 007137599bc4d69ee952a5228b5fcb2c3b9a95a7, and correlate endpoint evidence with traffic to 2[.]26[.]252[.]48:8080 or gate[.]dfwioeiofwr[.]info. Its reported Microsoft-CryptoAPI/10.0 user agent has changed across versions.
On a matched Windows host, isolate and preserve evidence, then inventory the AI, Git, SSH and cloud identities used there. Revoke exposed sessions and rotate affected keys from a clean machine. AWS recommends CloudTrail monitoring for access-key usage; GitHub says revoked tokens stop authenticating Git and API requests. Confirm old AWS keys are inactive in IAM, review their CloudTrail usage during the exposure window, and inspect GitHub access and the applicable oauth_authorization.destroy security-log event. Review each service reachable by the workstation’s credentials for activity during the exposure window.