The user wants a concise HTML summary of the provided news about LLM safeguards. The news is from arXiv, title: "Safeguards Based on Copyable Context Cannot Provide Reliable Safety for LLMs". Authors: Pingyu Wu, Lingyao Zhu, Weiming Zhang, Nenghai Yu. Date: 2026-07-29. Description: "Large language model safeguards decide whether to answer before seeing how an answer will be used. This creates a basic problem for dual-use tasks: the same answer can help an authorized professional or an attacker, while an attacker can imitate a benign request and interaction history. We separate the capability released by the model from the evidence available about downstream use. When that evidence is copyable, we derive the exact worst-case floor on attacker assistance while preserving usef" I need to write a 2-4 sentence summary in

and then a link to the original source. The summary must be in English, concise, technical, no invented info. The description cuts off at "preserving usef" likely "usefulness". I'll base summary on provided content. Summary: The paper argues that LLM safeguards relying on copyable context cannot reliably prevent misuse because attackers can mimic benign requests. It formalizes the separation between model capabilities and downstream-use evidence, showing that when evidence is copyable, there is a fundamental worst-case lower bound on attacker