HDM-2 is a novel system that detects and validates LLM-generated content against both context and common knowledge to prevent misinformation in enterprise settings.

• Introduces a taxonomy of enterprise LLM responses categorizing statements as context-based, common knowledge, enterprise-specific, or innocuous
• Employs a dual verification approach checking statements against provided context and general facts
• Achieves significant detection accuracy for hallucinations in business applications
• Offers practical deployment pathway for enterprise security teams

This research addresses critical security concerns by preventing AI systems from confidently presenting false information that could lead to business risk, reputational damage, or decision errors.

HalluciNot: Hallucination Detection Through Context and Common Knowledge Verification