A Multi-Agent Simulation of AI Compute Regulation, Compliance, and Deterrence.

This project models the strategic interaction between AI Labs (seeking to maximize profit by training models) and an Auditor (seeking to enforce compute permit limits). It serves as a computational playground to explore the conditions under which regulation succeeds or fails.

The simulator provides an interactive web dashboard for real-time experimentation.

• Python 3.13+
• uv (recommended for dependency management)

# 1. Install uv (if needed)
curl -LsSf https://astral.sh/uv/install.sh | sh

# 2. Clone and Sync
git clone https://gitlab.com/your-org/aisc-cm-simulator.git
cd aisc-cm-simulator
uv sync

Launch the visualization dashboard to explore the model interactively.

uv run solara run app.py

Key Features:

• Scenario Control: Adjust penalty ($P$), detection prob ($p$), and audit capacity ($N$) on the fly.
• Real-time Analysis: Watch compliance rates and market prices evolve.
• Agent Inspector: Drill down into individual lab behaviors and audit history.

Run headless simulations for bulk data collection.

uv run main.py

We use uv for all development tasks to ensure reproducibility.

This project includes a GitLab CI/CD pipeline that:

1. Tests: Runs pytest on every commit.
2. Deploys: Builds the Solara app (WASM) to GitLab Pages on merge to main.

For deep technical details on the architecture, decision logic, and mesa model structure, see: 👉 Technical Documentation