ResearchBenchmarksSecurity Profile Generation on Security profiling scenarios Sections 3.2.1–3.2.8, 3.2.10, and 3.2.11Follow75Direct Decision Match Rate (%)Claude59.463.4567.571.55Apr 7, 2026Evaluation ResultsMethodMethodLinksDirect Decision Match Rate (%)Regulatory RelevanceDecision Class AdequacyParameter AdequacyRationale QualityCorrection NeedControls Minor Edits (%)Hallucination Rate (Lacking Details) (%)ClaudeModel version=Sonnet 4...Model version=Sonnet 4.6, Inference setting=Extended2026.0475——————10ChatGPTModel version=Thinking...Model version=Thinking 5.4, Inference setting=Extended effort2026.0465——————25GeminiModel version=3.1 ProModel version=3.1 Pro2026.0460——————43