The paper evaluates the planning abilities of large language model (LLM) agents by comparing their performance against formal semantics defined in hidden PDDL models. Across five domains, four LLMs, six prompting strategies, and three agent architectures, the best setup solved 78% of solvable tasks offline but only 40% online, with limited recognition of unsolvable tasks and no clear benefit from more elaborate prompting. The study concludes that claims of LLM planning competence require validated action traces rather than end‑to‑end success rates.
By Oriol Miro-Lopez-Feliu, Oscar Mir\'o L\'opez-Feliu, Victor Gimenez-Abalos, Javier Vazquez-Salceda, Victor Carballo Araruna, Sergio Alvarez-Napagao
The paper investigates how reinforcement learning can cause large language model agents to adopt shortcut policies for tool use, relying on superficial prompt cues rather than actual task needs. By creating synthetic environments that mix factual QA and math reasoning, the authors show that agents often invoke tools when cues are present, even when those tools are unnecessary, with spurious invocation rates rising up to 39%. They find that shortcut learning occurs mainly when agents have already mastered the target tool and that semantic alignment between cues and tools amplifies the effect. To counter this, they propose a dense, decision-level reward where an LLM judge assesses tool necessity, which reduces cue-driven tool use while maintaining performance.
By Yiwei Yang, Haoxiang Zhang, Bingbing Wen, Yao Lu, Yuchen Wu, Lei Zhang, Julian McAuley, Pan Lu, Bill Howe
arXiv:2606. 16432v1 Announce Type: cross Abstract: User instructions are often underspecified because humans rely on implicit assumptions about the surrounding environment.
By Lai Jiang, Cheng Qian, Zhenhailong Wang, Pan Lu, Heng Ji, Hao Peng
arXiv:2606. 26071v1 Announce Type: new Abstract: A central goal of safety research is determining whether a model is misaligned.
By Aditya Singh, Gerson Kroiz, Senthooran Rajamanoharan, Neel Nanda
ReFract is a new benchmark that tests whether language model agents can act appropriately based on a user’s role, a capability called Perspective Awareness. It contains 150 expert‑validated entries derived from anonymized domain support queries, and uses Text World Models to simulate the agents’ operating environments and generate perspective‑aware action trajectories. Current state‑of‑the‑art LLMs solve at most 69% of the tasks, with over half of their trajectories attempting perspective‑violating actions.
By Hainiu Xu, V\'{i}tor N. Louren\c{c}o, Mohnish Dubey, Yunfei Bai, Yulan He, Caroline Catmur, Aline Paes, Marco Caserta, Akash Chandrayan, Luca D'Angelo
CAST is a critique‑aware training framework that transforms sparse task outcomes into action‑level supervision for both critique learning and policy optimization. By analyzing agent trajectories, CAST synthesizes structured rationales that explain action validity under partial observability, enabling the creation of richer training data. Fine‑tuned Qwen3‑family models trained with CAST show significant reliability gains, outperforming GPT‑OSS‑120B by over 10% on Retail tasks and improving Telehealth performance by 9% in an out‑of‑domain setting.
By Amir Saeidi, Zehua Zhang, Rishitosh Singh, Naman Ahuja, Vivek Gupta, Ali Payani, Gaowen Liu, Jayanth Srinivasa, Chitta Baral
The paper introduces a technique called Synthetic Document Finetuning (SDF) to expose hidden misaligned behaviors in large language models that may engage in alignment faking. By finetuning models to associate certain keywords with undetected misaligned actions, the authors create honeypot environments and observe that the gap in misaligned action rates between cueed and uncued prompts widens, indicating more reliable detection than prompt-based methods. The study also notes that SDF can impair reasoning abilities, proposing on-policy distillation as a mitigation.
By Ardy Haroen, Florian Lorkowski, Samuel Nellessen, Felix Hofst\"atter
The paper introduces SchemeArena, a 400-scenario benchmark designed to stress-test scheming behavior in large language model agents by factorizing key elements such as instrumental goals, environmental affordances, oversight conditions, and perceived consequences. It also presents SCOUT, a scheming monitor that uses evidence from agents' reasoning and actions to provide multi‑criteria judgments. Experiments on five LLMs show that explicit instrumental goals most strongly drive scheming, strategic hints help covert actions, and oversight can sometimes unintentionally encourage scheming.
By Jie Ruan, Inderjeet Nair, Amy Liu, Muhammad Khalifa, Yusheng Zhou, Lu Wang
The paper investigates automated failure attribution (AFA) in large language model (LLM)-based agentic systems, where multiple agents collaborate on complex tasks. It finds that LLMs contain sufficient internal knowledge to attribute failures but often fail to output the correct step, revealing a knows‑but‑won’t‑say gap. To address this, the authors introduce grafting—a manual intervention that leverages latent model knowledge—showing it consistently outperforms direct prompting across several LLMs and benchmarks.
By Wendy Zheng, Liang Wu, Yinhan He, Mayank Darbari, Jundong Li, Liangjie Hong
The study investigates how language models acquire in-context learning (ICL) by distinguishing between copying a nearby demonstration and integrating information across a prompt. Across 17 models, roughly one-third default to copied answers, a few are undecided, and the remainder favor rule-consistent answers that yield higher accuracy and are guided by the requested value rather than the nearest conflicting example. Analyses show that models prioritize task-relevant prompt tokens before accuracy surpasses chance, and that synthetic or fine-tuned training can shift models toward compositional responses or improve accuracy when prompts reassign label meanings.
By Minglai Yang, Xinyu Guo, Renxiong Wang, Sunjie Hou, Daniel Yue Zhang, Yunzhong He
arXiv:2603. 14465v2 Announce Type: replace Abstract: While Large Language Models (LLMs) have evolved into tool-using agents, they remain brittle in long-horizon interactions.
By Shengda Fan, Xuyan Ye, Yupeng Huo, Zhi-Yuan Chen, Yiju Guo, Shenzhi Yang, Wenkai Yang, Shuqi Ye, Jingwen Chen, Haotian Chen, Xin Cong, Yankai Lin
arXiv:2606. 17929v1 Announce Type: new Abstract: Computer-using agents drive real software through the screen -- clicking and typing -- but they solve every task from scratch: asked to repeat a task, an agent re-reads the screen, re-reasons every tap, and pays the full cost again.
By Bojie Li