Trustworthiness is the practical ability of an AI system to be used safely and predictably. It is not just “trust” in a vague sense; it combines reliability, security, accountability, transparency, robustness, privacy, and fairness. In cyber security, trustworthiness matters because AI is now embedded in decisions, automation, and security workflows, so a weak system can create real operational risk.
In attacks, poor trustworthiness shows up as poisoned training data, adversarial prompts, hidden model changes, or outputs that fail under unusual inputs. In defense, organizations improve trustworthiness by documenting data and model choices, testing for failures and abuse, limiting who can change the system, monitoring drift, and keeping humans responsible for high-impact decisions. Frameworks such as the NIST AI RMF treat trustworthiness as a lifecycle issue: the system must be governed, measured, and reviewed continuously, not assumed safe after deployment.