On April 2, the AI Security Lab of Ant Group and Tsinghua University jointly open-sourced ClawAegis - the first security defense plugin covering the entire lifecycle of OpenClaw, aimed at systematically addressing the security and reliability risks it faces, providing native protection capabilities for autonomous agents that are multi-dimensional, lightweight, and capable of immediate intervention.

As open-source intelligent agent frameworks such as OpenClaw become increasingly popular, the security risks hidden behind their automation capabilities are becoming more evident. From skill poisoning and memory data contamination, to being maliciously guided in intent, executing high-risk operations, or exhausting resources, intelligent agents face challenges throughout their entire lifecycle, including initialization, user input, model reasoning, intelligent decision-making, and service execution.

image.png

In response to these risks, ClawAegis has built a comprehensive, multi-layered defense system covering the above five stages, which can actively identify and intercept various runtime threats, including malicious instruction injection, sensitive information theft, user intent tampering, and unauthorized access.

As a lightweight security plugin built-in, ClawAegis can be seamlessly integrated into the OpenClaw framework, dynamically activating protective mechanisms at key execution nodes. At the same time, it provides highly configurable risk identification and handling strategies, making it easy for security operators to flexibly respond to different attack scenarios; for ordinary users, it offers transparent protection of sensitive files and Skill assets.

This open-source initiative follows the recent efforts by the Ant AI Security Lab to help fix multiple high-risk vulnerabilities in OpenClaw. In the future, both parties will continue to iterate on ClawAegis capabilities, working with the community to build a trusted, controllable, and traceable intelligent agent runtime environment.