#LLM Agents

Paper Review — An Era Where Safety and Security Come First

As generative models/LLMs are used in the real world, the “evaluation design” for safety and security becomes dominant. A cross-review of fresh papers on DESPITE, MAGIC, Claudini, and the limits of...

Paper Review - Latest Evaluation Designs Connecting Scheduling and Safety

Three new papers published in 2026-05〜06 on safety evaluation, agent behavior, and the limitations of long-form models. A shared theme is that realistic validation design strongly affects risk esti...

Paper Review - “Experience Compression” and Safe Operation of LLM Agents

Organizes, based on three recent arXiv papers on LLM agents, the framework for compressing experience to enable long-term execution, as well as emerging trends in safety evaluation and verification...