Agent Threat Rules Hit 113 Detections and AGT v3.2 Locks Down the Wire Layer

Tools

ATR v2.0.1 Ships 113 Detection Rules With 10/10 OWASP Agentic Coverage

ATR is becoming the Sigma ruleset for AI agents, and v2.0.1 now covers every OWASP Agentic Top 10 risk with 113 rules at 99.7% precision. Cisco AI Defense is already consuming these rules, which means this is production-grade signal, not a research toy.

AGT v3.2 Adds Signal Protocol E2E Encryption and AgentMesh Wire Protocol v1.0

The toolkit now ships full end-to-end encryption using X3DH key agreement and Double Ratchet across both Python and TypeScript SDKs, plus a published wire spec covering envelope format, cipher suite, and relay semantics. If you are building multi-agent systems that communicate across trust boundaries, this is the reference architecture to read.

AGT v3.5 Ships Cost Governance, Collective Policies, and Decision Lineage Reconstruction

v3.5 adds tiered cost budgets with auto-throttle and kill switches, multi-agent collective policy evaluation, and a Decision BOM that reconstructs the full lineage of any agent action from observability signals. The Citadel Entra identity bridge and APIM policy fragment round out enterprise deployment needs.

Analysis

How Microsoft's Agent Governance Toolkit Maps to Every OWASP Agentic AI Risk

The blog post lays out the full seven-package governance stack and explicitly maps each OWASP risk to a specific toolkit capability, from semantic intent classification for goal hijacking to ring isolation for rogue agents. Worth bookmarking as a reference architecture checklist before shipping any autonomous agent to production.