- Engineer — Plan: Infostealers harvesting AI API keys and session tokens from developer machines and CI/CD environments is a real exposure vector; audit all AI service credentials (Google, Anthropic, etc.) in your pipelines, rotate long-lived API keys, and enforce short token TTLs where providers allow it.
- SOC/IR — Plan: Lumma and Vidar are well-established infostealer families with known detection signatures; build or tune endpoint detections for these stealers specifically to flag AI service token harvesting, and add a hunt for anomalous AI API calls originating from unusual geolocations or IPs in recent logs.
- Leader — Plan: As enterprise AI tool adoption grows, stolen replayable tokens become a meaningful account-takeover vector that sidesteps MFA; use this quarter to inventory which AI platforms your org uses, establish an API key governance policy, and confirm vendor support for token revocation and audit logging.