AI Agent SecurityJune 2026·45 min read read
The Firebreak Stack
Security Engineering for AI Coding Agents
Eight controls for containing coding agents before bad tool calls reach production.
Key topics
- ›Why guardrails fail at the wrong abstraction layer
- ›The Firebreak Stack — eight operational gates
- ›Devcontainer reachability and scoped credentials
- ›Egress proxy, stratified confirmation, and budgets
- ›Deploy weekend, control plane, and proving each gate