BlockBeats news, on February 6, GoPlus Security launched GoPlus AgentGuard, a native security skill for AI agents, aiming to address execution-level security risks that have emerged after the proliferation of AI agents.
As AI agents begin to possess real execution capabilities (such as command execution, file reading and writing, service invocation, and on-chain transactions and signatures), the risk has evolved from "incorrect answers" to execution risks that could directly cause system damage, data leaks, or financial losses. Based on its long-term technical expertise in Web3 on-chain security and transaction risk control, GoPlus proactively moves security capabilities forward to before AI execution occurs.
GoPlus AgentGuard uses a "runtime interception + on-demand deep scanning" mechanism to intercept, confirm, or allow high-risk behaviors, assessing action risks before the AI Agent executes. This is a dedicated execution-level security mechanism to prevent AI Agents from "executing recklessly, executing beyond authority, or executing dangerously," allowing users to safely and confidently use OpenClaw. AgentGuard is now open source and available for use by AI Builders and developers.