Grounded but Misleading: Evaluating Semantic Alignment in AI-Generated Security Explanations
Online scams increasingly leverage fluent and context-aware social-engineering strategies, creating growing demand for AI systems that explain why a message may be risky. However, explanations that cite detector-derived evidence may still semantically weaken or redirect the intended risk interpretation. We introduce VEXA (Verifying Semantic EXplanation Alignment), a controlled testbed for studying the gap between lexical grounding and semantic risk alignment in AI-generated scam-risk explanations. VEXA generates ungrounded, risk-aligned, and risk-diluting explanations by independently controlling evidence grounding and semantic framing. Across LLM-as-a-judge and human evaluations, explanations can remain comparatively grounded even when their interpretations weaken the detector’s intended risk assessment. In human evaluation, risk-diluting XAI-grounded explanations retained comparatively elevated Perceived Evidence Grounding scores (3.66±1.02) despite lower Helpfulness (3.00±1.41) and Reasoning Support (3.14 ± 1.05) scores. These findings provide controlled evidence of grounding-illusion effects and suggest that trustworthy explanation evaluation must verify not only whether evidence is cited, but also how it is interpreted.
Publications
Heajun An, Connor Ng, Sandesh Sharma Dulal, Junghwan Kim, Jin-Hee Cho “Grounded but Misleading: Evaluating Semantic Alignment in AI-Generated Security Explanations.” In Proceedings of the 2026 AAAI/ACM Conference on AI, Ethics, and Society (AIES-26)