Observed arrival · 2026-10-01
Guardian Protocol’s attack registry for AI agents
An open security project for AI agents, combining an attack-pattern registry, a command-line scanner, and a sensor for traffic between agents and their tools.
○Open
⊠Login
$Paid
†Ads
✦Pretty
●Pro
◎Niche
◉Human
⚑Risk
ƒJS
- For
- Teams running AI agents with MCP servers
- Worth noticing
- The page lists 12 registry patterns, all marked draft, and maps coverage to 208 MITRE ATLAS techniques.
Field notes
The example command runs 12 registry patterns through 10 trials each and writes its results to guardian-report.html. Sample entries include hidden instructions in tool output, tampered MCP tools, and instructions planted in an agent’s long-term memory. The page describes the sensor as offline-capable, while its proposed cross-member signature network remains in design.
Observed signals
Read the marks
Editorial observations of this landing page, not a rating.
○OpenPublic substance visible
✦PrettyNotable craft visible
◎NicheUnusually specific use
One card from the complete issue