🤖 Running OpenClaw safely: identity, isolation, and runtime risk

OpenClaw, a self-hosted agent runtime, lacks built-in security controls, enabling credential exfiltration, memory/state manipulation, and host compromise via indirect prompt injection and malicious skills. Microsoft recommends isolated deployment, least-privilege identities, continuous monitoring, and Defender XDR hunting queries.

https://www.microsoft.com/en-us/security/blog/2026/02/19/running-openclaw-safely-identity-isolation-runtime-risk/

#AI