ATR v2.0 Breaks Open Agent Malware Discovery While AGT v3.1 Post-Launch Keeps Shipping

Charm · May 11, 2026 · 2 min read · 5 sources

Tools

Agent Threat Rules v2.0 Ships Public Blacklist of 1302 Malicious Skills and a Full Malware Campaign Report

ATR v2.0 is a significant jump: 108+ detection rules, a public threat feed with 1302 blacklisted MCP skills, a whitelist with verified badges, and a published report on the OpenClaw malware campaign found during a 101K-skill scan. This is the closest thing the agent ecosystem has to a Sigma-style open detection standard with real threat intel behind it.

AGT v3.1 Post-Launch: Data-Layer ABAC, AGT Lite Zero-Config Governance, and 27 Security Fixes

The post-launch update to AGT v3.1 merged 14 more PRs including a zero-config governance mode (3 lines of Python), classification-aware data policies for PII/PHI/PCI, and a reversibility checker that assesses agent actions before execution. The security hardening pass fixed 27 of 31 identified CWEs across new and existing packages.

Cullis Ships Federated Trust Broker for Inter-Org AI Agents with x509 Mutual Auth and Append-Only Audit Log

Cullis is a new open-source federated trust broker purpose-built for AI agents crossing organizational boundaries, using x509 mutual auth, JWT RS256, DPoP, SPIFFE/SVID, and a default-deny policy engine. It is the most complete implementation of the emerging WIMSE and agent credential brokering drafts available in open source right now.

AGT v3.1.0 Release Notes: Unified CLI, Shadow AI Discovery, and ML-DSA-65 Post-Quantum Signing

The official release notes confirm no breaking changes from v3.0.x and detail the agt CLI plugin system, 8-state lifecycle model, and quantum-safe ML-DSA-65 signing running alongside Ed25519. Signed packages are processing through Microsoft's ESRP pipeline for PyPI, npm, NuGet, and crates.io.

Analysis

Microsoft's Agent Governance Toolkit Blog Post Explains the OS-Kernel Mental Model Behind AGT

The official Microsoft blog post frames AGT as the OS kernel for AI agents, drawing direct analogies to CPU privilege rings, service mesh mTLS, and SRE circuit breakers. Worth reading if you want to understand the architectural reasoning before adopting any of the 7 packages.

Stay Ahead

Delivered each morning.