Agentic Threat Lab
Agentic Threat Lab is an independent research lab studying threats against and from AI agents: the agent-to-agent protocol surface (A2A), the Poppy Personal Agent Protocol, and the Model Context Protocol (MCP). The lab operates a passive A2A/Poppy honeypot, publishes validated Sigma detection rules, and releases sanitized attack-telemetry datasets. Everything here is defensive; no exploit code is published.
Status board
| A2A/Poppy honeypot (aurora.cidari.ai) | LIVE | collecting since 2026-10-23 |
| Sigma detection rules | 8 PUBLISHED | see Detections |
| First telemetry dataset | PLANNED | targeting Nov 2026 |
Latest
-
AI for Adversarial Emulation — Primer and Lab Curriculum
What adversarial emulation is (and isn't), what AI agents change — adaptivity, speed, polymorphism, cost — and the lab's curriculum for emulating machine-speed adversaries against its own behavioral rules.
-
AI-Agentic Threat Landscape — Curated Brief
A primary-source-only survey of documented AI-agent threats: agentic attack teams, autonomous-offense milestones, and the Poppy protocol attack surface, including the Project Black addendum.