Cisco open sources Project Quarry to detect rogue AI agents
Photo: Tima Miroshnichenko / Pexels
Quarry flags risky sequences across an entire agent session, such as a sensitive file read followed by an external send.
Cisco has open-sourced Project Quarry, a behavior detection engine that monitors AI agents so security teams can catch them going off the rails, The Register reported. Cisco wants it to provide a “shared language” for agent behavior, as Snort did for malicious network traffic and YARA did for malware.
Most detection tools look at single interactions, such as a prompt injection attempt. Quarry instead evaluates the whole agent session, flagging sequences like a sensitive file read followed by a request to an external endpoint. “What is missing is the entire context under which something may occur,” said Amy Chang, Cisco’s head of AI threat and security research.
Rules modeled on Snort and YARA
Quarry rules resemble Snort and YARA rules, with a header, content to match and metadata. The GitHub repository includes plugins for Claude Code, Codex and other MCP-compatible agents, plus a starter rule pack. Cisco said rules are not a substitute for human oversight.
Disclosure: This article was edited by Diego Almada Lopez. For more information on how we create and review content, see our Editorial Policy.

