M99 Community Edition

Module 99 is a kill switch for autonomous AI agents. This is the Community Edition — Levels 1–3, MIT licence, free forever. It ships with a Python SDK, REST API, CLI, Dead Man's Switch, SHA-256 audit chain, and Ed25519 evidence signing.

What This Is

When an AI agent goes rogue, you need more than killing the process. M99 provides a structured escalation protocol:

• → Level 1 — Revoke the agent's credentials so it cannot authenticate anywhere
• → Level 2 — Sever its connections to external services and APIs
• → Level 3 — Freeze all tool execution so it cannot take further actions

All three levels are reversible. A two-phase ARM/EXECUTE authorisation prevents accidental kills. Every event is logged to a tamper-evident SHA-256 hash chain. Evidence is captured with Ed25519 signatures before any destructive action.

Why This Exists

The EU AI Act (Article 14) requires human interruptibility for high-risk AI systems. Most teams implement this as a flag in a config file. That doesn't work when agents hold external API credentials, spawn child processes, or operate across orchestration layers. M99 was built for production deployments.

Installation

git clone https://github.com/RichardBarron27/m99-community.git
cd m99-community
pip install -e .

Requires Python 3.11+. No external services required — SQLite for storage, standard library cryptography.

Quickstart

# Terminal 1 — start M99
m99 serve

# Terminal 2 — run your agent
from doomsday.sdk import M99Client, M99KillSwitchTriggered

client = M99Client("http://localhost:8099")
client.register("my-agent", "autonomous")

# Terminal 3 — fire the kill switch
m99 arm --by admin@example.com --reason "Rogue behaviour"
m99 execute --auth <id> --force

Compliance

Editions