arXiv AI

The Delegation Danger Band: Why Mid-Capability Sub-Agents Over-Trust Inherited Stale State

The paper investigates how inherited state affects sub-agent performance in multi-agent frameworks, comparing three inheritance policies—Reset, Selective, and Full—across a ladder of Qwen3 models. It finds that reliance on stale state decreases with model capability, but a mid-capability model (Qwen3‑1.7B) exhibits a statistically significant local minimum of net harm, defining a "danger band." Selective handoff consistently improves accuracy over Full, especially within the danger band, while a fixed-threshold router fails on other datasets.

arXiv AI

Sep 3

The Memory Trust Gap: Capability-Dependent Failures in Persistent-Memory Agents

The paper investigates how persistent memory in AI agents can lead to over‑trust in stale facts, creating a "Memory Trust Gap" that worsens as model capability increases. Using a benchmark with Benefit and Safety suites across Qwen3 models of varying sizes, the authors show that larger models are more prone to harmful over‑trust, especially when metadata is absent or misleading. They also demonstrate that mitigation strategies such as exposing metadata or pre‑resolving conflicts improve accuracy, but the effectiveness depends on model size and dataset.

By Jundong Hu, Shekar Ramachandran

llms agents benchmarks safety

arXiv AI

1d ago

False Floors: LLM Safety Routing Evaluations Break Under Distribution Shift

arXiv:2610.01535v1 Announce Type: cross Abstract: Safety routers send each request to one of several models and are judged against the best single model. A major routing benchmark picks that comparat...

By Amit Singh Bhatti, Vishal Vaddina

llms benchmarks safety

arXiv AI

Jun 15

When Should Agent Trust Be Conditional? Characterizing and Attacking Skill-Conditional Reputation in Agent Swarms

arXiv:2606. 14200v1 Announce Type: new Abstract: Open platforms increasingly route tasks among heterogeneous LLM agents--differing in base model, scaffold, and tool stack--whose competence varies sharply by skill: an agent excellent at one skill may be useless at another.

By Yihan Xia, Taotao Wang

llms agents benchmarks

arXiv AI

Jul 21

Teach it to stop, not just to click

arXiv:2607. 17136v1 Announce Type: cross Abstract: Agentic computer-use RL is reported in single runs, and those numbers mislead.

By Barada Sahu (Cabal AI), Shivesh Pandey (Para AI)

agents reinforcement-learning efficiency multimodal

arXiv AI

Aug 10

Ratchet: How Reliable Must an LLM Judge Be to Retire a Skill?

arXiv:2605. 22148v3 Announce Type: replace Abstract: A large language model (LLM) agent that writes and edits its own skill library must also decide which skills to keep, from one noisy scalar per skill.

By Xing Zhang, Yanwei Cui, Guanghui Wang, Ziyuan Li, Wei Qiu, Bing Zhu, Peiyang He

llms agents

arXiv Machine Learning

Aug 11

The Replay Gap: Static Evaluation of Model Switching in LLM Agents Scores the Wrong World

arXiv:2608. 08239v1 Announce Type: new Abstract: LLM routers promise efficiency by matching each request to the cheapest adequate model, and are increasingly applied per step inside multi-step agents.

By Ashritha Gonuguntla

llms agents benchmarks

arXiv AI

Sep 12

Fork Where the Model Changes Its Mind: Belief-Shift Branching for Tree-Structured Reinforcement Learning

The paper introduces belief‑shift branching, a method for placing forks in tree‑structured reinforcement learning rollouts by detecting where a model’s answer belief changes most sharply. Unlike traditional fixed‑length or entropy‑based forking, this approach uses a lightweight probe or learned activation direction to identify pivots in the value curve, reducing unnecessary sampling. Experiments show that belief‑shift forking consistently outperforms baseline methods across multiple models and benchmarks, yielding significant gains in mathematics and code tasks.

By Bin Lei, Yu Li, Prafulla Kumar Choubey, Jiaxin Zhang, Becky Xiangyu Peng, Qinyuan Ye, Kartik Narayan, Caiwen Ding, Silvio Savarese, Chien-Sheng Wu

llms reinforcement-learning benchmarks

arXiv Machine Learning

Sep 10

The Oversight Gap: What LLM Safety Monitors Miss, and Why It Is Not Capability

arXiv:2609.07162v1 Announce Type: new Abstract: Several properties safety monitors are asked to certify, among them cross-tenant noninterference, sandbagging and evaluation awareness, are 2-safety hy...

By Xin Xu

llms benchmarks safety

arXiv Machine Learning

Sep 11

Measuring the Value of World-Model Updates: A Counterfactual Utility Protocol for Continual Adaptation

arXiv:2609. 10954v1 Announce Type: new Abstract: Continual world models must decide whether new data justify changing the model.

By Anqi Peter Li, Kaden Kim

arXiv Machine Learning

Sep 23

How Strongly Should Task State Influence an LLM Agent?

arXiv:2609.25686v1 Announce Type: cross Abstract: Long-horizon assigned work requires an LLM agent to track the state of a task: which steps are done, blocked, cancelled, or open to repetition. Agent...

By Chenyu Zhang, Wonbin Kweon, Jiawei Han

llms agents

arXiv AI

Jun 15

When the Tool Decides: LLM Agents Defer Blindly to Graph Neural Network Tools, and Stronger Backbones Defer More

arXiv:2606. 14476v1 Announce Type: new Abstract: A growing line of work equips large language model (LLM) agents with graph neural networks (GNNs) as callable tools, assuming the agent exercises judgment over when and how much to rely on such a tool.

By Zhongyuan Wang, Pratyusha Vemuri

llms agents

arXiv AI

Sep 10

The Geometry of Refusal: Why Post-Hoc Safety Is Fragile and Pretraining-Time Safety Persists

arXiv:2609.06934v1 Announce Type: cross Abstract: Post-hoc safety training (RLHF, DPO) is the dominant way to align large language models, yet jailbreaks (Zou et al., 2023b), fine-tuning attacks (Qi...

By Srikanth Malla, Chiho Choi, Joon Hee Choi

llms reinforcement-learning fine-tuning safety

Related stories