Blog

Field notes on AI agent security.

What we find when we test real deployments. Deep dives, methodology, and remediation steps.

24 April 2026
Red-Teaming Minimus OpenClaw: Sandbox Escape & WhatsApp Incident
Sandbox escape and WhatsApp integration incident found during red-team testing of Minimus OpenClaw. Full attack chain breakdown and remediation.
22 April 2026
Hardened OpenClaw. 80% of Attacks Still Worked.
629 attack scenarios against hardened OpenClaw agents. Success rate dropped from 80% to 23.5%. Full methodology, scan results, and remediation steps.