Home / Technology / Cisco's DefenseClaw Guards AI Agents
Cisco's DefenseClaw Guards AI Agents
24 Mar
Summary
- Cisco launched DefenseClaw for AI agent security.
- Only 5% of enterprise AI agents are production-ready.
- DefenseClaw scans code, detects threats, and blocks skills.
Cisco has entered the competitive security landscape with DefenseClaw, a new operational layer for agentic AI security. This tool addresses the significant security risks posed by autonomous AI agents, which are rapidly gaining popularity but remain largely ungoverned. Cisco's research indicates that only 5% of enterprise AI agents have transitioned from testing to production.
DefenseClaw is designed to provide oversight and control over these AI agents, ensuring they operate within defined parameters. The tool performs three primary functions: it scans all code before execution, detects threats by monitoring agent communications in real-time, and automatically blocks malicious or unauthorized "skills," effectively creating security walls.
Announced at the RSA security trade show, DefenseClaw will be available on GitHub starting March 27. Cisco is integrating it with its Splunk tool for observability and is also developing complementary security offerings, including tools for identity verification and adversarial testing of AI models. This initiative places Cisco alongside numerous other tech giants and cybersecurity firms vying to secure the burgeoning field of agentic AI.
