Web Intellectual Property at Risk: Preventing Unauthorized Real-Time Retrieval by Large Language Models
A semantic defense framework that embeds optimized HTML policy cues to prevent unauthorized real-time LLM retrieval — supporting refusal, partial masking, and source redirection — lifting defense success rates from 2.5% to 88.6% across multiple proprietary LLMs.
Jan 2025
