# Cisco open-sources Project Quarry to detect threats across multi-step AI agent behavior

> Original title: An AI agent's actions may look harmless individually. The real risk emerges when they're connected.

- Source: X search: AI launch posts (introducing, just launched)
- Published: 2026-10-28T12:58:37.000Z
- Bad Influence: https://badinfluence.news/items/k3i2xc4i1ervf1ntb60s3schf
- Original: https://x.com/CiscoAI/status/2115428019901002214

## Summary

Cisco introduces Project Quarry, an open-source detection language for identifying threats across multi-step AI agent behavior. The post says individual agent actions may look harmless, while risk emerges when those actions are connected.

## Post

An AI agent's actions may look harmless individually. The real risk emerges when they're connected.

Introducing Project Quarry: @Cisco's new open source detection language designed to identify threats across multi-step agent behavior.

See how it works: https://blogs.cisco.com/ai/project-quarry-agent-security-detections

![](https://badinfluence.news/api/img-proxy?u=https%3A%2F%2Fpbs.twimg.com%2Fmedia%2FHVuCkpnbYAAOosb.jpg&mode=thumb&exp=1793404800&sig=a12c4c190aa888ce)