Robodebt has shown that relying on automated decision making tools can be disastrous. What checks and balances could be developed to guide the safe and ethical use of AI in government operations, policy making and service delivery?