Beyond Exact-Match: Semantic Authentication for AI-Native Wireless Systems

AI-native NextG wireless systems increasingly exchange compressed, generated, and machine-interpretable representations whose value lies in their meaning rather than in exact message recovery. This shift creates a mismatch with conventional authentication mechanisms, which verify bit-level or message-level equality and may reject benign semantic transformations. We propose a semantic authentication framework that accepts a recovered meaning whenever it remains within an admissible equivalence class of the intended meaning. The equivalence classes are induced by a description-length-based semantic distance, which measures the excess coding cost incurred when an observation generated under one meaning is interpreted under another. Using this metric, we develop a set-based authentication rule with an invariant semantic hash, analyze the resulting false-rejection probability under benign perturbations, and validate the analytical results through simulation.