- March 8, 2026
We built a test corpus for AI agent egress security tools
72 attack cases across 8 categories. Secret exfiltration, prompt injection, MCP tool poisoning, chain detection. Any security tool can run against it. No vendor lock-in.
- March 5, 2026
Guardrails deleted, now what?
OBLITERATUS and similar tools remove safety guardrails from open-weight models using weight ablation. When the model won't refuse, your only defense is the network layer.
- February 21, 2026
What is an agent firewall?
AI agents make HTTP requests, call tools, and handle credentials. An agent firewall scans traffic in both directions before anything gets through.
- February 14, 2026
EU AI Act Runtime Security: What You Need Before August
The EU AI Act's high-risk requirements take effect August 2, 2026. The compliance standard won't land until Q4. Here's what to build now if you're running AI agents.
- February 13, 2026
The First AI Agent Espionage Campaign: What Defenses Matter
Anthropic disclosed GTG-1002, the first AI agent espionage campaign. A state actor jailbroke Claude Code for autonomous hacking. What happened and which defenses work.