GoPlus launches AgentGuard to build an execution-level security boundary for OpenClaw

Foresight News reported that GoPlus Security has launched a security Skill called "AgentGuard," specifically designed for AI Agents with real-world execution capabilities, aiming to provide execution-level security protection for OpenClaw.

As AI Agents begin to possess real execution abilities (such as command execution, file reading and writing, service invocation, and on-chain transactions and signatures), the risks have escalated from "incorrect answers" to execution risks that could directly cause system damage, data leakage, or financial loss. Leveraging its long-term technical expertise in Web3 on-chain security and transaction risk control, GoPlus has shifted security capabilities to precede AI execution.

GoPlus AgentGuard uses a "runtime interception + on-demand deep scanning" mechanism to intercept, confirm, or allow high-risk behaviors, assessing action risks before AI Agents execute them. This is a dedicated execution-level security mechanism to prevent AI Agents from "arbitrary execution, unauthorized execution, or dangerous execution," enabling users to use OpenClaw more safely. AgentGuard is now open source and available for AI Builders and developers.