Safe enablement for autonomous agents