Resilience
An AI system's ability to recover from failures, adapt to changing conditions, and maintain acceptable operation under stress.
Definition
Resilience in AI systems refers to the ability to anticipate, withstand, recover from, and adapt to adverse conditions, attacks, or failures while maintaining acceptable levels of operation. It encompasses both technical robustness and organizational adaptability.
Regulatory Context: The EU AI Act addresses resilience through its robustness and cybersecurity requirements. Article 15 requires high-risk AI systems to be resilient against errors and inconsistencies, while Article 15(4) specifically requires resilience against unauthorized third-party attempts to alter use or performance.
Dimensions of AI Resilience:
- Technical resilience: System continues functioning despite component failures or attacks
- Operational resilience: Ability to maintain service during disruptions
- Adaptive resilience: Capacity to evolve in response to changing conditions
- Recovery capability: Speed and completeness of return to normal operation
Key Requirements:
- Redundancy and failover mechanisms
- Graceful degradation under stress
- Error detection and correction
- Backup and recovery procedures
- Incident response capabilities
Regulatory Standards:
- EU: AI Act Article 15; Digital Operational Resilience Act (DORA) for financial sector
- US: NIST AI RMF; sector-specific resilience requirements
- ISO: ISO 22316 (organizational resilience); ISO/IEC 27031 (ICT readiness)
Related concepts: Robustness, Reliability, Cybersecurity Requirements, Risk Management System
Sources
- •EU AI Act Article 15
- •NIST Cybersecurity Framework
Related Terms
Robustness
The ability of an AI system to maintain acceptable performance and resist failures, attacks, or unexpected conditions across its lifecycle....
Reliability
The ability of an AI system to perform consistently and correctly under specified conditions over time....
Cybersecurity Requirements
Technical and organisational measures that ensure AI systems resist, detect, respond to, and recover from cyber threats across their lifecycle....
Risk Management System
Continuous process to identify, analyze, and mitigate AI system risks throughout its lifecycle....