As Artificial Intelligence AI becomes integral to next-generation networks, balancing performance with accountability, resilience, and privacy is critical. This paper introduces an iterative AI/ML development framework that systematically integrates trustworthiness metrics alongside traditional utility measures. By incorporating eXplainable AI (XAI) techniques, this methodology ensures models are not only accurate but also interpretable and robust against adversarial threats. This framework has been validated through two main use cases in networking and security. In encrypted traffic classification, results demonstrate that while Gradient Boosting achieves higher raw performance, Random Forest provides a superior balance between utility ( \(94.3\%\) ) and accountability. In malware detection, we show that strategic feature reduction yields minimal utility degradation while substantially improving resilience to evasion attacks by \(15\text {-}85\%\) . Unlike existing approaches addressing isolated aspects of security, this framework provides a comprehensive, quantifiable methodology for managing trade-offs, supporting the deployment of trustworthy AI systems within modern, high-stakes communication infrastructures and network environments.