Capture Notes

Paper on relinking risks at the compression boundary of LLM agents.

AI security relevance:

• Relevant to summarization, memory compression, context pruning, and cross-session state safety.
• Potentially important for GraphRAG and long-context agent pipelines.