Fail-Safe

A design principle ensuring that an AI system defaults to a safe state or ceases operation when a failure, error or anomalous condition is detected.

Designing AI to default to a safe state when things go wrong. If a self-driving car's AI can't figure out what to do, it should pull over and stop; not speed up.

Fail-safe design is a governance principle that should be applied to all AI systems. Your framework should require that AI systems default to a safe state when errors, failures or anomalous conditions are detected.