Blocking an AI agent is not the same as stopping it
Results from 180 coding-agent runs on how policy communication changes compliance, and why a clean final workspace can hide protected reads, temporary copies, and retained backups.
Read article →Self-Governance Is Not Governance: The Case for Out-of-Process Enforcement
A practical analysis of trust boundaries in agent governance and why layered, out-of-process policy enforcement is required for production safety.
Read article →Why Your AI Agent's Built-In Safety Needs an Independent Enforcement Layer
A case study on the Claude Code deny-rule bypass, Project Glasswing context, and why independent runtime enforcement is required for defense in depth.
Read article →Do AI Agents Actually Respect Guardrails?
Observations from testing agents inside a live policy enforcement layer, including where agents comply, where they route around constraints, and what deployment teams should lock down first.
Read article →Configuring AI Agent Security Is Harder Than It Should Be
Policy controls are scattered, scopes overlap, and documentation assumes you already know what you are doing. Here is how to make sense of it.
Read article →