GuardClaw is a security and compliance control plane for agentic AI computer use that sits between agents and OpenClaw/MCP servers. CISOs and security engineers are increasingly worried on X/HN about agents exfiltrating data, clicking the wrong thing, or violating SOC2/ISO controls, while current “AI security” tools mostly focus on prompts, not OS-level behavior. GuardClaw introduces policy-as-code for what agents can click, read, and execute across desktops, browsers, and internal tools. - Least-privilege policies for MCP/OpenClaw: define which skills, hosts, and file paths are permitted per task, user, or model. - Real-time guardrails: block risky actions (e.g., external uploads, privilege escalation, mass deletes) with human-in-the-loop approvals. - Compliance reporting: exportable audit trails mapping agent actions to SOC2/ISO/NIST controls for faster audits. - Vendor-neutral integration: plugins for popular agent frameworks plus SIEM connectors (Splunk, Datadog, Panther). Monetization is via enterprise subscriptions (per-seat + per-host) targeting mid-market and enterprise orgs piloting AI agents in finance, healthcare, and SaaS.