ExAI5G: A Logic-Based Explainable AI Framework for Intrusion Detection in 5G Networks

Transparency doesn't require sacrificing performance: ExAI5G achieved 99.9% accuracy while maintaining interpretability, directly challenging the common assumption that explainable AI systems must accept lower performance than "black-box" models.

Logical rules make AI decisions actionable for operators: The framework extracted 16 interpretable logical rules with 99.7% fidelity to the underlying model, enabling security teams to understand why the system flagged a threat and respond appropriately rather than blindly trusting a prediction.

Multiple explanation techniques strengthen trustworthiness: By combining Integrated Gradients (for feature importance), decision trees (for logical rules), and LLM-generated explanations, the framework provides complementary perspectives on model reasoning, reducing reliance on any single explanation method.

LLM-generated explanations can be rigorously evaluated: The paper introduces a novel methodology using evaluator LLMs to assess whether AI explanations are both "faithful" (accurate to the model's actual reasoning) and "actionable" (useful for decision-making), suggesting that AI explanations shouldn't be accepted at face value without validation.

Explainability is critical infrastructure for high-stakes security: In 5G network intrusion detection, operators need to understand and trust automated decisions that protect critical infrastructure—making interpretability not just a nice-to-have feature but a requirement for responsible deployment.