A technical requirement under the EU AI Act for high-risk AI systems, demanding that they maintain an appropriate level of resilience against both accidental and malicious attempts to tamper with their performance, such as through the input of adversarial examples.

‍