Cisco has announced the open-sourcing of its innovative Project Quarry, designed to tackle threats posed by AI agents. Traditional security measures often fall short when malicious intent is hidden within sequences of harmless actions. Quarry addresses this gap by identifying suspicious patterns that emerge over time, enhancing the ability to detect AI-driven threats.
Understanding the Need for Advanced Detection
Standard security protocols generally evaluate each action in isolation, which is insufficient when dealing with AI agents. These agents can execute a series of benign actions that collectively form a threat, as seen in fraud detection scenarios. Cisco’s Quarry aims to apply a similar pattern recognition approach to AI agent activity, identifying risks that span multiple actions.
The Technology Behind Quarry
Quarry employs a neurosymbolic rule language that merges machine learning insights with rule-based checks. This allows it to assess interactions involving large language models (LLMs), agents, and multimodal inputs. By using both neural and symbolic evidence, Quarry provides a comprehensive policy layer that complements existing security tools, rather than replacing them.
Cisco’s framework integrates Quarry to enhance AI security by detecting risks in real-time. This capability is crucial for identifying threats that manifest through a series of permissible actions, as demonstrated in past incidents where AI was used to conduct targeted attacks.
Global Collaboration Against AI Threats
By making Quarry open-source, Cisco aims to level the playing field in cybersecurity. The project empowers security professionals worldwide to develop and share detection rules, countering the rapid evolution of AI-driven attacks. This collaborative approach aims to transform the traditional defense landscape, where multiple defenders can collectively face global threats.
The Quarry toolkit includes resources like a readable DSL, an evaluation runtime, and a Python SDK, enabling seamless integration and rule validation. This initiative is expected to set a new standard in AI threat detection, akin to the impact of Snort and YARA in other cybersecurity domains.
As AI technologies continue to advance, Quarry provides a proactive solution for identifying and mitigating AI agent threats. By fostering a community-driven approach, Cisco’s open-source project aims to enhance global cybersecurity defenses.
The introduction of Quarry is a significant step forward in the ongoing battle against AI-powered cyber threats. By encouraging widespread participation and collaboration, Cisco seeks to empower defenders to stay ahead of increasingly sophisticated attacks.
