Draft regulations splits AI applications into risk-based tiers and bans some