AgentSnare is a trajectory‑adaptive deception system that expands a decoy environment while autonomous web‑penetration agents explore. It uses an LLM‑driven decoy constructor and a virtual kernel to generate and materialize decoy artifacts that persist across interactions, enabling long‑horizon exploration. In experiments on 15 CVE‑Bench web applications and four attacker models, AgentSnare delays 49.9% of tool calls, diverts 57.2% of post‑entry actions into decoys, and defuses 89.5% of completion reports, preventing any real target exploitation at pass@3 across 60 attacker‑CVE pairs.
By Ruoyu Wang, Heng Zhao, Renjie Wu, Tianhang Zheng, Mengnan Zhao, Wanyu Lin, Kui Ren
CuriousTogether is a framework that evaluates curiosity-driven interactions among multiple large language models (LLMs) across novelty-seeking, epistemic, and social curiosity dimensions. In eight tasks covering reasoning, creative generation, and social bias mitigation, curiosity-guided multi-agent settings outperform non-curious ones on seven of the eight tasks, achieving up to a 20.6% increase in story semantic diversity and a 14.9 percentage point drop in stereotype rates. The study also shows that country-conditioned cultural grounding can improve story generation and social tasks, though it may reduce precision in reasoning, and that inference-time curiosity-based multi-agent approaches offer better quality-token trade-offs compared to single-agent baselines.
By Angana Borah, Zoha Hayat Bhatti, Zhijing Jin, Rada Mihalcea
ShiftRAG is a new retrieval‑augmented generation framework that decouples discriminative retrieval learning from autoregressive generation while still using a single LLM backbone. It addresses the textual bottleneck of existing RAG pipelines by projecting retrieval alignments into continuous soft tokens, creating a high‑bandwidth interface that feeds retrieval signals into the generator. The approach also introduces a soft orthogonality objective to reduce rank collapse and layer‑wise relevance dynamics to preserve fine‑grained latent space, resulting in significant improvements in retrieval and question‑answering performance over state‑of‑the‑art baselines.
By Youmin Ko, Jihong Jeong, Hyunjoon Kim
The paper examines how fine‑tuned Transformer detectors (DistilBERT, RoBERTa) perform on an out‑of‑distribution dataset of hyper‑local news generated by LLMs. Using SHAP‑based explainable AI, the authors uncover three phenomena: a Call‑to‑Action bias, a Specificity Bias where hallucinated named entities evade detection, and a Wikipedia Paradox where formal human prose is misclassified as AI. These results highlight systematic shortcomings of purely statistical detectors in structured human domains and underscore the need for semantically‑aware, interpretable detection methods.
By Nico Morloher, Johannes Reschke, Mohamed Mady
The paper introduces a comprehensive audit protocol for System One decision models, which provide typed, probability‑scored answers without human oversight. Using three adapted benchmarks—BBQ, CrowS‑Pairs, and SilverSpoon—it evaluates twelve models plus a no‑thinking baseline across 1,638,000 decisions, revealing widespread stereotype alignment, order‑dependent responses, and confidence instability. The study demonstrates that swapping answer options and reversing question order significantly alters model outputs, underscoring the need for order‑invariant scoring in procurement.
By Alejandro Mozo Quesada (Universidad de M\'alaga, Spain), Sergio G\'alvez Rojas (Universidad de M\'alaga, Spain), Francisco L\'opez Valverde (Universidad de M\'alaga, Spain)
The paper investigates whether disagreement among user simulators can flag unreliable estimates of group response proportions and group differences. Using six language‑model simulators on 60 questions across 12 countries, disagreement more accurately predicts invalidity for group differences (AUC 0.87) than for single group proportions (AUC 0.71). The authors find that shared bias among simulators can obscure errors, but correcting for this bias reduces prediction error without altering disagreement signals.
By Saber Zerhoudi
The paper introduces VIAS, a method for Value-Isolating Activation Steering that aims to improve multi-value compliance in large language model agents. VIAS constructs steering vectors by contrasting responses that include all relevant values with those that omit a single target value, ensuring the target value changes from compliant to non-compliant while maintaining compliance on other values. Evaluations on the PROTECT workplace compliance dataset show that VIAS reduces vector overlap and, when combined with prompting, achieves higher compliance rates across three models compared to prompting alone.
By Avni Mittal, Avinash Anand, Dikshant Kukreja, Erik Cambria, Rajiv Ratn Shah, Sandipan Dandapat
The paper introduces ARCBench, a configurable benchmark for measuring rule compliance in language‑model agents that compact long histories to fit context windows. Using ARCBench, the authors find that rule violations after compaction rarely stem from rule deletion; instead, agents often ignore the rule while it remains present in the compacted history. The study shows that the source of violations depends on how the rule is delivered and on the compaction strategy, with targeted compactor attacks significantly increasing violations, and that simply retaining a rule in the compacted summary does not guarantee compliance.
By Yiwei Zha, Qingyu Meng, Herbert Bos, Min Chen
arXiv:2610.35254v2 Announce Type: replace
Abstract: The number of paper submissions to major AI conferences has increased dramatically in recent years, substantially increasing the peer-review worklo...
By Sai Suresh Macharla Vasu, Sumedh Santosh Joshi, Philipp Christmann, Ivaxi Sheth, Ruta Binkyte, Koel Dutta Chowdhury, Raouf Kerkouche, Hui-Po Wang, Mario Fritz
arXiv:2607.28347v2 Announce Type: replace
Abstract: LLMs are increasingly deployed as proxies for human study participants in social science experiments, yet the fidelity of this practice has rarely...
By Sebastian Pohl, Harsh Mehta, Pranav Mambayil, Abdul Ghafoor, Franziska Lesigang, Yufang Hou, Christian Hilbe
arXiv:2604.16656v2 Announce Type: replace
Abstract: All languages are equal; when it comes to tokenization, some are more equal than others. Tokens are the hidden currency that dictates the cost and...
By Maitrey Mehta, Nishant Subramani, Zhichao Xu, Ashim Gupta, Vivek Srikumar
The paper introduces SYNC, a training‑free inference controller that dynamically selects the optimal layer for injecting persona steering vectors during decoding. SYNC estimates each layer’s sensitivity to every persona using cosine similarity trajectories, allowing the model to adaptively choose the best injection point at each step. Experiments on two models and multiple persona profiles show that SYNC outperforms static multi‑persona steering methods in 21 of 28 configurations, with significant gains in 15 cases.
By Yosef Nuraga Wicaksana, Muhammad Reza Qorib, Fadhil Muhammad, Alfan Farizki Wicaksono
The paper introduces a KL regularization term that aligns predicted dialogue act distributions with empirical transition patterns derived from a corpus. Applied to a 60‑class German counselling taxonomy, this method improves macro‑F1 scores by 9–42% relative to various encoders and enhances dialogue‑flow alignment. Cross‑dataset validation on the HOPE dataset shows that the gains transfer across languages and counselling domains, and systematic ablations reveal that transition regularization consistently benefits weaker baseline models, especially in fine‑grained, data‑sparse dialogue tasks.
By Eric Rudolph, Philipp Steigerwald, Jens Albrecht
EffComp is a new framework for prompt compression in retrieval‑augmented question answering. It uses BERT‑style document rerankers and sentence selectors to perform fast, extractive compression at the sentence level, thereby avoiding hallucinations. The hybrid reinforcement‑plus‑supervised training optimizes compression ratio while maintaining LLM accuracy, and experiments show EffComp outperforms existing methods with low latency.
By Romtham Tungsununtham, Piyawat Lertvittayakumjorn, Atiwong Suchato, Peerapon Vateekul
The paper introduces a continual learning approach for large language models called Multi‑Teacher On‑Policy Distillation (MOPD) that avoids storing intermediate expert snapshots by replaying prompts from earlier domains using the previous student as a teacher. Two replay strategies are examined: a shared replay budget across domains and a fixed budget per domain, with replay steps that either distill token distributions alone or also align hidden states, attention scores, and value relations. Experiments on a five‑domain benchmark (logic, math, code, safety, instruction following) demonstrate that replay alone closes most of the performance gap between sequential distillation and joint MOPD, achieving comparable results to recovery from all stored snapshots when using a fixed replay share per domain.
By Yiqiao Qiu
The paper introduces CLEO (Constrained Loop for Environment Optimization), a technique for tailoring a minimal agent harness around a self-hostable large language model (LLM) to meet the needs of government intelligence teams. CLEO selectively incorporates tools, guardrails, memory policies, and loop components that enhance accuracy while keeping the system general and auditable. On a held‑out dataset, the CLEO harness achieves an 83.8% pass rate—just 1.2 points below the top out‑of‑the‑box Claude Code SDK baseline—yet is two orders of magnitude smaller.
By Shai Nahum Gefen, Vered Chassida, Dmitry Basin, Guy Feigenblat
The paper introduces latent debate, a structured surrogate framework that interprets the internal thinking of Large Language Models (LLMs) on True/False tasks by modeling latent arguments and their interactions. It provides a model- and task-agnostic conceptual framework, instantiated symbolically to approximate LLM decision-making. Empirical results show that latent debate closely matches LLM predictions, offers interpretability, and serves as a strong baseline for hallucination detection, revealing correlations between debate patterns and hallucination risk.
By Lihu Chen, Xiang Yin, Francesca Toni
arXiv:2611.03092v1 Announce Type: cross
Abstract: Personal AI agents are increasingly capable of handling long-horizon tasks across applications, yet users often need to invest substantial effort in...
By Zheng Wu, Xinbei Ma, Yao Yao, Zhaoyuan Huang, Haowen Hu, Yijie Lu, Taolei Wang, Yifei Yang, Wuyue Xiong, Pengzhou Cheng, Zhuosheng Zhang
The paper introduces a grammar‑lens benchmark that generates data at progressively higher computational levels to test language models. Using this benchmark, GPT‑5.5 and its Pro variants were evaluated, showing increasing acceptance rates of 16%, 40%, 43%, and 52% as the models’ test‑time reasoning levels rose. The study also notes that higher compute effort leads to responses that diverge more from human answers, hinting at a potential capability‑versus‑alignment tension.
By Emma Graham, Richard Granger
TopoSteer introduces a topology-based method for selecting the layer at which to apply activation steering in large language models. Instead of evaluating separability or alignment, it scores layers by the total H_0 persistence of the displacement cloud, measured as the length of its Euclidean minimum spanning tree. Experiments across several models and tasks show that this approach yields more effective steering without needing held-out validation.
By Tuc Nguyen, Marcos Guti\'errez-del-Pozo, Eduardo Paluzo-Hidalgo, Yifan Hu, Adaku Uchendu, Thai Le