Observed incident
WardenStealer collects AI-agent credentials and local context
Reported
Published here
What happened
Gen reported AI-tool targeting by CallbackBeaver, later identified as WardenStealer. SpyCloud’s October 30 analysis of recovered logs found Claude and Codex credentials and configuration files collected from infected endpoints.
Parties and reporting
AI tools and systems: WardenStealer / CallbackBeaver, Claude Code and Desktop, Codex CLI
Outcome
Observed collection extends beyond malware-advertisement claims. Downstream use of particular stolen AI credentials and the number of AI users affected remain unknown.
Operating lesson
After endpoint compromise, revoke AI tokens and sessions, rotate reusable MCP secrets and review linked-service access.
Evidence and limits
September 8 is the earliest reviewed AI-targeting disclosure; October 8 resolved the family name. SpyCloud’s total family-infection count is not an AI-account compromise count. Malware must already execute on the endpoint; no new AI-product flaw is established.
Sources
- Gen initial AI-agent targeting disclosure
Primary source ·
- Gen Warden attribution and technical analysis
Primary source ·
- SpyCloud recovered-log evidence
Primary source ·