Egypt National Guidelines for Trustworthy and Responsible AI (Edition 2.0, March 2026) 3.2b: Pre-deployment TEVV: validate performance, robustness, security and safety, fairness and ethics, and data quality and privacy

What does 3.2b require?

Model performance is verified against approved thresholds for accuracy, latency, robustness, scalability and uncertainty under normal conditions, at the edges and under stress, with graceful degradation and reproducibility across environments; the system is tested for how it withstands noise, perturbation and unexpected input, how it resists adversarial and cyber attack, and whether its fail-safe, fallback and safe-mode behaviours work; differences in impact and in performance between demographic and contextual groups are assessed, with ethical and societal risks and trade-offs documented against organisational values and external frameworks; and data completeness, accuracy, representativeness, lineage and bias controls are verified with privacy by design, data minimisation, retention and lawful processing confirmed, testing with real-world and synthetic datasets to expose hidden risks.