AI systems should be monitored and tested to ensure they continue to meet their intended purpose, and any identified problems should be addressed with ongoing risk management as appropriate.