The ability of an AI system to continue operating correctly even when some components fail or produce errors.
Involves architectural patterns - redundant components, graceful degradation, checkpointing, and transaction rollbacks - that ensure AI services remain available and safe under partial failures. Governance prescribes fault-injection testing (chaos engineering), failure-mode analyses, and clear service-level objectives for recovery time and service continuity.
A cloud-based image-classification API runs on multiple container instances behind a load balancer. If one instance crashes during a high-volume event, traffic automatically shifts to healthy pods, and crashed instances restart without user impact. The ops team regularly performs chaos tests to validate fault-tolerance mechanisms.
