I Spent the Last Few Days Testing AI Agents and Got Scared — So I Built Sentinel v0.3.0

Sentinel v0.3.0 was released as a security layer for AI agents to reduce jailbreaks, tool abuse, and data leaks.

Its core design moves controls outside the agent into an independent Shield Sidecar that can approve risky actions.

The system also adds deterministic shadow sandboxing, a red-team engine, and monitoring/compliance features.

The goal is stronger auditability and external, out-of-band safeguards for safer agent deployment, including EU AI Act readiness.