Trustworthy AI refers to AI systems that are reliable, secure, explainable, and appropriate for the task they are meant to do. In practice, this means the model behaves consistently, its outputs can be reviewed, and its limits are understood before it is used in production or in high-stakes decisions.
In cyber security, trustworthiness matters because AI often handles sensitive data, automates workflows, or connects to internal tools. Attackers may try prompt injection, data poisoning, or malicious tool instructions to steer a model into unsafe actions or data leakage. Defenders reduce this risk with access controls, input filtering, retrieval and tool isolation, logging, model monitoring, and human review. Trustworthy AI is not a single feature; it is a set of controls that helps keep AI useful without making it a new attack surface.