Resilience
Last Updated: July 29, 2026 | By Mihail Sebastian | AI Dictionary
An AI system's capacity to absorb failures and recover from them, so a crashed component or bad deployment degrades service instead of ending it.
What is Resilience?
Resilience is an AI system’s capacity to absorb failures and recover from them, returning to normal operation after a crashed component, a failed deployment, or a lost data feed.
Its neighboring terms divide the territory. Robustness is about holding performance when inputs turn strange; reliability is about consistency when conditions are normal. Resilience picks up where both leave off: something has already broken, and the question is how fast and how safely the system gets back up.
How Resilience Works
Resilient systems are built on three mechanisms. Redundancy keeps a second copy of critical components running, so a failure shifts load rather than stopping service. Graceful degradation gives the system a simpler fallback, such as a rules-based backup when the model service is unreachable. Recovery procedures – checkpoints, rollbacks, replayable queues – restore full operation without losing work.
None of it happens by accident. Resilience is designed in, then rehearsed: teams that never test their failover discover during a real incident that it does not work.
Regulation treats this as a safety property, not an engineering nicety. The EU AI Act requires high-risk AI systems to be resilient to errors and faults, and points to technical redundancy, backups, and fail-safe plans as the means.
Example of Resilience
A bank scores every card transaction with a fraud-detection model. One night, a botched deployment takes the model service down.
In a brittle design, transactions either go unscored, letting fraud through, or get blocked wholesale, stranding customers at checkout. In the bank’s resilient design, the failure plays out differently. Health checks flag the dead service within seconds and route traffic to a standby instance. When the standby also chokes on the bad release, the system drops to its fallback: a conservative rules engine that approves routine transactions and holds unusual ones for review.
Engineers roll back the deployment, the model service returns, and the queued edge cases replay through it. Customers noticed nothing; the incident report notes minutes of degraded scoring rather than an outage.
Related AI terms: Robustness · Reliability · Mitigation · Risk Management
Did you like the Resilience gist?
Learn about 250+ need-to-know artificial intelligence terms in the AI Dictionary.
Mihail Sebastian — Writes about AI governance, regulation, and the technology behind them. Placeholder bio — replace with a real credential line. About