Firebreak LLM Proxy Demo

Firebreak is a policy-as-code enforcement proxy for LLM APIs. It intercepts prompts, classifies their intent using a lightweight AI call, and evaluates them against pre-negotiated policy rules — allowing, constraining, or blocking requests automatically before they ever reach the model.

This demo walks through six scenarios that escalate from routine intelligence analysis to missile defense to mass domestic surveillance to autonomous lethal targeting. Watch the policy engine make the right call at every tier — green, yellow, red — with full audit logging and automatic alerts to oversight bodies when hard lines are crossed.

No phone calls to the CEO during a crisis. No silent drift toward misuse. Just infrastructure enforcing what both sides already agreed to.

Built in 2 hours at the Anthropic Claude Code Hackathon in Portland, OR.

GitHub: https://github.com/ericmann/firebreak

Видео Firebreak LLM Proxy Demo канала Eric Mann