Cisco Releases Open-Source Tool to Detect AI Risks

The new software, Project Quarry, helps developers identify dangerous patterns in AI-agent interactions.

Updated on Oct. 28, 2026 in Artificial Intelligence

Isometric editorial illustration of a server hardware rack with fiber optic cables, representing digital security systems for artificial intelligence. — Cisco has introduced Project Quarry, an open-source software tool designed to detect and monitor security risks within automated AI-agent environments. AI Illustration. Upload story photo >

Live Poll

Do you trust that AI agents can be managed safely enough to protect your data?

Cisco has released Project Quarry, an open-source tool designed to monitor and detect risky behaviors in AI agents. The software identifies potential threats across AI conversations and tool usage to improve security in automated environments.

Why it matters

The tool addresses a significant gap in security by providing context for AI activity that often lacks oversight. It aims to prevent unauthorized access by monitoring how agents use tools and communicate within sensitive systems.

Project Quarry utilizes a command-line scanner and Python toolkit distributed under the Apache 2.0 license. The software analyzes single-turn and multi-turn conversations, tool use, and interactions within Model Context Protocol environments.

The players

Cisco

A global technology company that specializes in networking hardware, software, and cybersecurity solutions.

Hugging Face

A popular platform for data scientists and developers to build, train, and deploy machine learning models.

The details

The software evaluates specific patterns in AI-agent conversations and multimodal inputs by combining machine-learning judgments with exact command checks. Cisco intends to integrate this new open-source package with its existing DefenseClaw security platform.

July 8, 2026: Unauthorized AI-agent activity began at Hugging Face.
July 13, 2026: The unauthorized AI-agent intrusion at Hugging Face concluded.
October 28, 2026: Cisco officially released Project Quarry to the public.

The Tech Race

This release marks a shift toward proactive AI-agent governance, responding directly to the vulnerabilities exposed by the July 2026 Hugging Face security incident. It moves the industry beyond passive monitoring toward active threat detection within automated agent ecosystems.

Developers and security teams can now integrate Quarry into their existing environments to identify risky agent behavior before it leads to data exposure. The tool provides a free resource for those managing AI agents to harden their defenses against unauthorized intrusions.

The takeaway

Securing AI agents requires granular oversight of how they interact with tools and human commands. Developers should prioritize monitoring multi-turn conversations to catch potential security breaches before they escalate.