🧰 LLM agents

The items that name it 5

• REALM: Identifying Minimal Task Assets for LLM Agents
  paper2026-10-29
• The One-Pixel of Reasoning: One Truthful Relation Is Enough to Mislead LLM Agents
  paper2026-10-17It is written here as LLM agents (eighteen models tested, proprietary and open-weight, on biomedical inference and vulnerability detection tasks).

  eighteen models tested, proprietary and open-weight, on biomedical inference and vulnerability detection tasks

• ASPIRE: Agentic Safety & Prompt Injection Red-teaming Engine
  paper2026-10-06

  Agents that retrieve untrusted content and act through tools, susceptible to indirect prompt-injection risks causing unauthorized actions or persistent state changes

• Semantic Behavioral Watermarking: Paraphrase-Robust and Forgery-Resistant Provenance for LLM Agents
  paper2026-10-06

  five agent models evaluated: Qwen2.5-3B and others across the 3B-14B range

• Closed-World Resolution Against Tool Hallucination in LLM Agents
  paper2026-09-16It is written here as LLM agents (tool-augmented).

  Ten hosted models tested under two invocation surfaces (raw-JSON and Model Context Protocol); a 675B-parameter model and a 7-8B-parameter model both exhibit tool hallucination