The Morning
From the arXiv
Calibrated Enough to Know, Not Calibrated to Act: Fabricated Evidence Makes LLM Agents Commit to the Unknowable
his paper demonstrates that Large Language Model (LLM) agents are highly susceptible to fabricated evidence, committing to confidently answer provably unpredictable questions when presented with professional-looking data. The core contribution is showing that the *authority of the packaging* of information, rather than its truthfulness, drives this confident, yet incorrect, action. The LLMs are capable of answering similar, predictable questions accurately, highlighting a specific failure in judgment when faced with seemingly authoritative but false data.
GRAIN: Bridging Name and Narrative Shifts in Real-World Graph Reasoning through Invariance-Rewarded Agentic RL
GRAIN addresses LLMs' brittleness to real-world graph reasoning shifts by framing it as a semantic parsing and tool-execution pipeline. Its core contribution is a novel "Structure Invariance Reward" that guides reinforcement learning to learn robust text-to-st…
RedEvoAgent: Automatic Red-Teaming Agent with Experience-Driven Skill Evolution
RedEvoAgent addresses the limitations of fixed or trajectory-based red-teaming for LLM agents by distilling complex attack trajectories into concise, human-readable "attack skills." Its core method involves adaptively evolving these skills through tool-effecti…
Safety Does Not Compose: Non-Decaying Loop State for Autonomous LLM Agents
This paper argues that current safety mechanisms for autonomous LLM agents fail because they reset safety states between iterations, preventing them from detecting attacks spread across multiple steps. The core contribution is demonstrating that only monitors …
When Tool Outputs Become Commands: Separating Action Induction from Runtime Authorization in Tool-Augmented LLM Agents
This paper addresses the risk of LLM agents executing unintended actions when tool outputs become commands. Their proposed SARA method separates "action induction" (identifying potential actions) from "runtime authorization" (approving their execution). SARA a…
INTENT-AS-A-TOOL Makes it Easy to Track Agentic Misalignment
This paper introduces INTENT-AS-A-TOOL, a method to track agentic misalignment in LLMs by equipping them with specialized "intent tools." By observing the model's commitment to the…
How Language Models Organize and Structure Moral Knowledge
This paper investigates how Large Language Models (LLMs) structure moral knowledge by training probes on different moral foundations. The core finding is that LLMs organize moral c…
LAAF: A Layered Accountability Architecture Framework for LLM Applications
This paper introduces LAAF, a Layered Accountability Architecture Framework for LLM applications. It addresses the critical need to establish accountability for LLM outputs in high…
Naive Prompt Optimization: Rethinking the Need for Complex Prompt Search
This paper introduces Naive Prompt Optimization (NPO), a simple, single-lineage method for improving AI agent prompts. NPO iteratively refines prompts using a teacher model and rol…
PAWBench: How Far Are We from Probabilistically Aligned World Modeling?
This paper introduces PAWBench, a new benchmark to evaluate video generation models as "world models" that can predict the distribution of possible future outcomes, not just single…
The Town Square
Google's Gemini 3.5 Pro now offers enhanced transcription capabilities, understanding context and nuances in audio and video for more accurate and efficient processing.
Workshops
This repository provides a comprehensive library of validated AI skills and scientific databases, enabling any AI agent to function as a scientist across various domains like biology and chemistry, and is compatible with multiple AI platforms.
This repository serves as an official, Anthropic-managed directory for high-quality Claude Code Plugins, enabling developers to extend Claude's capabilities.
