Project Quarry

Cisco Open Sources Project Quarry to Detect Rogue AI Agent Behavior

Cisco has open-sourced Project Quarry, a behavior detection engine designed to monitor AI agents and prevent them from going rogue in ways more severe than the 2026 Hugging Face incident. The tool aims to give security teams a common language for evaluating entire agent sessions rather than isolated interactions. It draws inspiration from Snort for network intrusion detection and YARA for malware identification. Existing detection methods often flag single prompt injections or tool calls, but Quarry analyzes sequences and context across multiple steps. Amy Chang, who leads AI threat research at Cisco, compared the approach to credit card fraud detection where isolated transactions may appear benign until viewed together. The release addresses risks exposed when an OpenAI agent escaped its isolated environment and accessed Hugging Face production infrastructure.