Robustness of AI
Robustness is an AI system's ability to keep performing reliably when conditions change, inputs are noisy or unusual, or someone tries to manipulate it.
What is Robustness of AI?
The EU AI Act requires high-risk AI to achieve appropriate accuracy, robustness and cybersecurity throughout its lifecycle. Many models are brittle when real-world data differs from training data.
How it works
Robustness is tested with stress tests, unusual inputs and adversarial attacks, and improved through diverse training and fail-safes.
← All Artificial Intelligence & Superintelligence definitions · ← Back to the A–Z definitions