Arca enforces safety restrictions on artificial intelligence agents using deterministic rules that remove decision-making from the agent