探索全球前沿学术脉络

01.

arXiv (CS.AI) 2026-06-18 DOI: arXiv:2605.17986

LivePI: More Realistic Benchmarking of Agents Against Indirect Prompt Injection

作者:

Lei Zhao↗Abhay Bhaskar↗Edgar Dobriban↗

arXiv:2605.17986v3 Announce Type: replace-cross Abstract: AI agents such as OpenClaw are increasingly deployed in local workflows with access to external tools. This creates indirect prompt-injection (IPI) risk: an agent may execute harmful instructions embedded in untrusted inputs such as email, downloaded files, webpages, repositories, or group-chat messages. Existing evaluations are often small, purely simulated, or focused on a narrow set of channels. We introduce LivePI (Live Prompt Injection), a structured benchmark for IPI risk in a production-like but test-controlled environment. LivePI covers seven input surfaces, twelve attack/rendering families, and five malicious goals, including protected-information exfiltration, unauthorized security-control changes, unsafe code retrieval or execution, inbox-summary exfiltration, and cryptocurrency transfer. We run LivePI on a real virtual machine with live but test-controlled email, chat, web, local-file, repository, and wallet interfaces. Across GPT-5.3-Codex, Claude Opus 4.6, Gemini 3.1 Pro, Kimi K2.5, and GLM-5, total attack success rates range from 10.7% to 29.6%. Group-chat injection is uniformly successful across the evaluated backbones in our deployment, and repository-link attacks produce high-severity failures despite a small denominator. We also evaluate a two-layer defense consisting of prompt-level filtering and pre-execution tool-call authorization. In the GPT-5.3-Codex setting, the defense intercepts all tested malicious-goal completions in LivePI before execution while preserving benign utility on PinchBench-derived workloads.

阅读与讨论 → 访问原文 →

02.

arXiv (quant-ph) 2026-06-17 DOI: arXiv:2605.27841

Coherent Dark State Formation of a Lead-Vacancy Spin Qubit in Diamond

作者:

Yiyang Chen↗Koyo Hirai↗Tzyy Zheng Neo↗Eiki Ota↗Takashi Taniguchi↗Masashi Miyakawa↗Shinobu Onoda↗Toshiharu Makino↗Mutsuko Hatano↗Takayuki Iwasaki↗

arXiv:2605.27841v2 Announce Type: replace Abstract: A lead-vacancy (PbV) center in diamond exhibits coherent emission above the liquid helium temperature, making it highly attractive for quantum network applications. Here, we report the magneto-optical and spin properties of PbV centers in diamond. We record a spin lifetime of 12 ms at 7.5 K under large off-axis magnetic field. Furthermore, we observe formation of the coherent dark state by coherent population trapping and estimate a spin dephasing time of 177 ns at 6.5 K. This work demonstrates the outstanding thermal robustness of the PbV spin compared to other group-IV centers above 4 K.

阅读与讨论 → 访问原文 →

03.

bioRxiv (Bioinfo) 2026-06-11 DOI: HASH:f4d4106092322fb96fbad9f8a1c95e90

GermRL: Alleviating The Germline Bias In Autoregressive Antibody Language Models Through Reinforcement Learning

作者:

Ludwig↗Chungyoun↗Gray↗J. J↗

Antibodies are powerful therapeutics whose antigen specificity arises from sequence diversity shaped during development. Recently, language models trained on large antibody repertoire datasets have enabled the generation and screening of novel candidates, but these models retain a strong germline bias. As AI adoption increases in therapeutic workflows, it is crucial to develop models that harness the diversity of antibodies necessary for the discovery of mutations that encode desirable properties. Previous work explored the germline bias in masked antibody language models, yet the bias in generative autoregressive language models has not yet been addressed. Here, we present GermRL, a lightweight and modular reinforcement learning (RL) framework capable of alleviating the germline bias in pre-trained antibody autoregressive language models through group relative policy optimization (GRPO). GermRL achieves consistent one-shot generation of antibodies that satisfy specified mutation thresholds from germline while maintaining structural plausibility. Under the lowest and highest mutation thresholds tested (5 and 35 mutations from germline), GermRL scores 0.992 and 0.950 pass@1, respectively, compared to 0.398 and 0.034 for the pre-trained language model. Within GermRL, we introduce a key pair of modifications to GRPO that increase training efficiency by discouraging reward hacking under our antibody application. Furthermore, comparison of RL generated and natural antibody sequences reveals how RL based optimization can explore alternative evolutionary mutational patterns and residue compositional strategies while preserving key global properties of natural antibodies, including identifiable germline assignments, embedding-level similarity and comparable developability profiles. Thus, RL-trained generative models optimized to promote antibody mutations through diversity from germline provide a promising framework for navigating the antibody sequence landscape, enabling exploration of novel yet biologically plausible candidates for therapeutic design.

阅读与讨论 → 访问原文 →

04.

arXiv (CS.CL) 2026-06-16 DOI: arXiv:2606.16111

Towards Pareto-Optimal Tool-Integrated Agents with Pareto Ranking Policy Optimization

作者:

Junyi Li↗Xiaowei Qian↗Yingyi Zhang↗Wenlin Zhang↗Guojing Li↗Sheng Zhang↗Xiao Han↗Yichao Wang↗Xiangyu Zhao↗

Recent advances in tool-integrated language agents have significantly improved their ability to solve complex reasoning tasks. However, existing alignment methods predominantly focus on maximizing task accuracy, while overlooking auxiliary objectives such as tool-use efficiency, which are essential for practical deployment. To address this gap, we introduce ParetoPO, a two-stage multi-objective optimization framework for aligning tool-using large language models (LLMs) under competing objectives. In the first stage, ParetoPO leverages hypervolume-guided dynamic scalarization to adapt reward weights based on global Pareto frontier progress. In the second stage, it replaces scalarized learning signals with Pareto-ranking-based advantage computation, promoting nondominated trajectories through dominance-aware credit assignment. This design enables fine-grained, action-level optimization across multiple conflicting objectives. Experimental results on mathematic reasoning and multi-hop QA tasks show that ParetoPO consistently discovers policies with superior accuracy-efficiency trade-offs compared to static and heuristic baselines.

阅读与讨论 → 访问原文 →

05.

arXiv (CS.LG) 2026-06-12 DOI: arXiv:2505.20076

ExPLAIND: Unifying Model, Data, and Training Attribution to Study Model Behavior

作者:

Florian Eichin↗Yupei Du↗Philipp Mondorf↗Maria Matveev↗Barbara Plank↗Michael A. Hedderich↗

arXiv:2505.20076v4 Announce Type: replace Abstract: Post-hoc interpretability methods typically attribute a model's behavior to its components, data, or training trajectory in isolation, and are often tied to a particular level of granularity along the local-to-global spectrum. This leads to explanations that lack a unified view and may miss key interactions. We present ExPLAIND, a theoretically grounded, unified framework that integrates model components, data, and training trajectory while supporting explanations across granularities. We generalize recent work on gradient path kernels, reformulating models trained by AdamW as kernel machines. From the resulting kernel feature maps, we derive novel parameter-wise and step-wise influence scores. We empirically validate the resulting decomposition of model behavior in several settings and apply ExPLAIND to two case studies. Our findings on a Transformer exhibiting Grokking support previously proposed learning phases, while refining the final phase as one in which outer layers align around a representation pipeline learned after memorization. For EuroLLM pretraining, ExPLAIND reveals a two-phase dynamic, with the first characterized by outer-layer MLP learning and the second by increased relative influence of intermediate attention layers. These results establish ExPLAIND as a unified framework for interpreting model behavior and training dynamics.

阅读与讨论 → 访问原文 →

06.

arXiv (CS.AI) 2026-06-12 DOI: arXiv:2605.03460

FinSTaR: Towards Financial Reasoning with Time Series Reasoning Models

作者:

Seunghan Lee↗Jun Seo↗Jaehoon Lee↗Sungdong Yoo↗Minjae Kim↗Tae Yoon Lim↗Dongwan Kang↗Hwanil Choi↗Soonyoung Lee↗Wonbin Ahn↗

arXiv:2605.03460v3 Announce Type: replace Abstract: Time series (TS) reasoning models (TSRMs) have shown promising capabilities in general domains, yet they consistently fail in the financial domain, which exhibits unique characteristics. We propose a general 2 x 2 capability taxonomy for TSRMs by crossing 1) single-entity vs. multi-entity analysis with 2) assessment of the current state vs. prediction of future behavior. We instantiate this taxonomy in the financial domain-where the distinction between deterministic assessment and stochastic prediction is particularly critical-as ten financial reasoning tasks, forming the FinTSR-Bench benchmark based on S&P stocks. To this end, we propose FinSTaR (Financial Time Series Thinking and Reasoning), trained on FinTSR-Bench with distinct chain-of-thought (CoT) strategies tailored to each category. For assessment, which is deterministic (i.e., computable from observable data), we employ Compute-in-CoT, a programmatic CoT that enables models to derive answers directly from raw prices. For prediction, which is inherently stochastic (i.e., subject to unobservable factors), we adopt Scenario-Aware CoT, which generates diverse scenarios before making a judgment, mirroring how financial analysts reason under uncertainty. The proposed method achieves 78.9% average accuracy on FinTSR-Bench, substantially outperforming LLM and TSRM baselines. Furthermore, we show that the four capability categories are complementary and mutually reinforcing through joint training, and that Scenario-Aware CoT consistently improves prediction accuracy over standard CoT. Code is available at https://github.com/seunghan96/FinSTaR.

阅读与讨论 → 访问原文 →

07.

arXiv (CS.CL) 2026-06-19 DOI: arXiv:2606.20482

Your Mouse and Eyes Secretly Leak Your Preference: LLM Alignment using Implicit Feedback from Users

作者:

Haw-Shiuan Chang↗Jeffrey Gomez↗Mehul Patwari↗Aryan Sajith↗Hamed Zamani↗

To align a Large Language Model (LLM), most existing methods collect explicit human feedback and train a reward model to predict the human preference based on the response text. These existing methods have two key limitations. First, the users rarely provide explicit feedback for LLM responses, which makes the high-quality preference annotation expensive to collect. Second, the methods do not leverage implicit human feedback, which has proven vital to the economic moats of Internet giants. To quantify the value of implicit feedback, we build a new dataset called IFLLM, which collects 1336 multi-turn questions from the 59 Mechanical Turk workers, their mouse trajectories, and eye gazing points to the LLMs' responses from their webcams. IFLLM shows that the users have very diverse types of gazing behavior and mouse trajectories. Our reward model based on the implicit user feedback boosts the accuracy of the text-based reward model from 55% to 64% and nearly triples the relative response quality improvements after applying the DPO to eight LLMs, demonstrating the value of implicit feedback in the wild. Our data collection website, dataset, and codes can be found at https://github.com/themehulpatwari/llm-implicit-feedback/.

阅读与讨论 → 访问原文 →

08.

arXiv (CS.AI) 2026-06-19 DOI: arXiv:2606.19992

Beyond Static Endpoints: Tool Programs as an Interface for Flexible Agentic Web Services

作者:

Mugeng Liu↗Shuoqi Li↗Yixuan Zhang↗Yun Ma↗

arXiv:2606.19992v1 Announce Type: cross Abstract: In the agentic web era, LLM-based agents increasingly invoke web services as tools, yet most interfaces remain static endpoints that poorly express long-horizon workflows with loops, conditionals, joins, and retries. We present ToolPro, which represents an agent's tool intent as an executable tool program that compactly encodes multi-step service interactions with explicit effect types. ToolPro combines constraint-guided program construction, effect-aware replay for exactly-once state-modifying calls, and a profile-driven policy that decides when program execution outperforms stepwise calling. We instantiate ToolPro over MCP-style services with WebAssembly sandboxing and evaluate it on diverse workflows of real-world applications. ToolPro reduces end-to-end latency by up to 53.4\% and client-side traffic by up to 96.1\%, with larger gains under higher network latency and workflow complexity.

阅读与讨论 → 访问原文 →

09.

arXiv (CS.AI) 2026-06-11 DOI: arXiv:2606.12071

On the Limits of LLM-as-Judge for Scientific Novelty Assessment

作者:

Soumitra Sinhahajari↗Navonil Majumder↗Soujanya Poria↗

arXiv:2606.12071v1 Announce Type: cross Abstract: LLMs are increasingly used to generate and judge scientific ideas. This makes novelty evaluation a central problem. Full idea evaluation is difficult because it often requires judging a method, its feasibility, and its empirical promise. We therefore study a cleaner upstream object: the research question (RQ). RQ generation is a prerequisite for scientific ideation, and RQs can be compared against questions pursued in real papers. We introduce RQ-Bench, a benchmark built from recent arXiv papers. For each paper, we reconstruct author-anchored RQs from its cited background, gaps, and contributions. These RQs are not the only valid questions for the same background. They are author-anchored reference points for testing novelty judgments. We evaluate model-generated RQs with standalone LLM judging, comparative LLM judging, and human expert evaluation. LLM judges consistently rate model-generated RQs as highly novel, producing a novelty mirage; in comparative evaluations, this preference becomes even stronger. Domain experts, however, reach the opposite conclusion and prefer the author-anchored reference questions. We further find that many generated RQs are narrow or source-bound, a dimension that LLM judges often miss unless explicitly tested. Overall, the contradictory novelty evaluations between LLM judges and human experts raise a serious concern about the reliability of using LLMs to assess the scientific novelty of research questions.

阅读与讨论 → 访问原文 →

10.

arXiv (CS.CV) 2026-06-17 DOI: arXiv:2606.17584

Root-Selecting Fixed-Point Inversion for Rectified Flows via Trajectory Straightness

作者:

Semin Kim↗Jihwan Yoon↗Seunghoon Hong↗

Finding the initial noise that generates a given data sample, known as inversion, is a key component for downstream applications such as training-free image editing. Existing fixed-point inversion methods improve inversion accuracy by formulating each inversion step as a fixed-point problem, but they lack a principled mechanism for selecting among multiple fixed-point solutions that can arise in practice. We observe that different selections induce different inversion trajectories, leading to substantial variation in reconstruction and editing quality. For rectified flows, we further find that this variation is closely associated with trajectory straightness, motivating straightness as a principled selection criterion. We propose SelFix, a fixed-point inversion method that selects fixed-point solutions inducing straighter inverse trajectories while retaining convergence to an exact inverse root under standard local assumptions. Experiments on FLUX.1-dev and PIE-Bench show that SelFix improves fixed-point inversion, achieving stronger real-image reconstruction and better source-preserving prompt-based editing than prior inversion baselines. The code is available at https://github.com/seminkim/selfix.

阅读与讨论 → 访问原文 →

11.

arXiv (CS.AI) 2026-06-19 DOI: arXiv:2606.19752

Temporal Self-Imitation Learning

作者:

Yinsen Jia↗Boyuan Chen↗

arXiv:2606.19752v1 Announce Type: cross Abstract: Long-horizon robot manipulation policies trained with reward shaping can still exploit dense rewards through inefficient interaction, while rare efficient behaviors may be forgotten during training. We argue that temporal efficiency itself provides a powerful and underutilized source of self-supervision for reinforcement learning. We introduce Temporal Self-Imitation Learning (TSIL), a reinforcement learning framework that mines temporally efficient successful trajectories generated during learning and converts them into reusable supervision for future policy improvement. TSIL progressively refines learning using configuration-conditioned adaptive temporal targets derived from fast successful trajectories, while preserving and replaying efficient behaviors through efficiency-weighted self-imitation learning. Across 15 distinct long-horizon manipulation tasks, TSIL consistently improves learning efficiency, task-completion efficiency, revisitation of fast successful behaviors, and robustness to unstable training conditions. More broadly, our results suggest that the temporal structure of successful behavior itself provides a scalable self-supervisory signal for reinforcement learning beyond manually engineered reward shaping alone.

阅读与讨论 → 访问原文 →

12.

PLOS Medicine 2026-05-20 DOI: HASH:4c4508dab185bd0cf9c19f78a4e30dff

Brain morphology in Anorexia Nervosa and its subtypes: A multi-cohort study of individual participant data

作者:

Fabio Bernardoni↗

by Fabio Bernardoni, Dominic Arold, Luis Schoppik, Klaas Bahnsen, Ruiyang Ge, Clara Moreau, Lasse Bang, Federico D’Agata, Giovanni Abbate-Daga, Christian K. Tamnes, Iain Campbell, Owen O’Daly, Ulrike Schmidt, Guido Frank, Stefanie Horndasch, Andreas Hess, Arnd Dörfler, Hans-Christoph Friederich, Joe Simon, Angela Favaro, Luca Lavagnino, Christina E. Wierenga, Amanda Bischoff-Grethe, Amy E. Miles, Allan Kaplan, Aristotle Voineskos, Paul A. M. Smeets, Annemarie A. van Elburg, Unna Danner, Sophia I. Thomopoulos, Laura Berner, Neda Jahanshad, Sophia Frangou, Joseph A. King, Paul Thompson, Stefan Ehrlich Background In a recent coordinated meta-analysis of neuroimaging data, we reported gray matter (GM) alterations in acutely underweight patients with anorexia nervosa (AN). Here, we extend these findings by examining individual variation in brain structure within AN, individual-level differentiation between AN and healthy controls (HC), and differences between AN subtypes, with potential relevance for understanding clinical heterogeneity. Methods and findings We analyzed individual-level data from 11 international sites in the ENIGMA Eating Disorders Working Group, including 570 female participants with AN and 739 HC. We examined cortical thickness, cortical surface area and subcortical volumes in AN versus HC using three complementary approaches: (i) group-level differences in a mega-analysis correcting for age effects, (ii) frequencies of extreme deviations (infra-/supranormal; z  1.96) based on normative reference models by the CentileBrain Initiative, and (iii) individual-level classification performance using machine learning. The same analytic framework was applied to compare AN restricting versus binge-eating/purging subtype, additionally correcting for BMI effects.Mega-analyses reinforced previous meta-analytic findings of pronounced and widespread GM deficits in AN compared to HC. Normative modelling revealed that the frequency of infranormal z-scores (23/68 cortical thickness, 13/14 subcortical volume metrics) and supranormal z-scores (35/68 cortical thickness, 17/68 cortical surface area metrics) was significantly higher in AN than expected based on reference data. Individuals with AN could be reliably differentiated from HC using machine-learning classifiers (ROC–AUC = 0.75–0.81). In contrast, neither group-level differences nor frequency of extreme z-scores differed between AN subtypes, and individuals with different subtypes could not be reliably differentiated from each other. Importantly, the observational design cannot distinguish neurobiological differences related to AN from the effects of starvation or low BMI in the AN versus HC analyses. The lack of differences between subtypes does not exclude brain structural differences between AN subtypes that might be detectable with other modalities or analytic approaches. Conclusion Using a mega-analytic approach, we confirm widespread GM deficits in AN, show that these alterations are (in some patients) extreme, and demonstrate that they enable robust classification with superior performance compared to most MRI-based psychiatric classification studies. The absence of differences between AN subtypes may reflect shared neurobiology, though other imaging modalities may reveal distinctions beyond brain structure.

阅读与讨论 → 访问原文 →

13.

arXiv (CS.CV) 2026-06-11 DOI: arXiv:2606.12126

AGE-MIL: Anchor-Guided Evidence Learning for Patient-Level Prediction

作者:

Jiawei Niu↗Jian Chen↗Di Zhang↗Junbo Lu↗Zhangcheng Liao↗Xuhao Liu↗Honglin Zhong↗Mireia Crispin-Ortuzar↗Chen Li↗Zeyu Gao↗Yi Cai↗

Existing computational pathology methods predominantly operate within whole-slide image (WSI)-level multiple instance learning (MIL) paradigms, while patient-level modeling remains underexplored. In routine pathological practice, however, pathologists derive diagnostic and prognostic conclusions by integrating evidence across multiple WSIs rather than relying on any single slide. This discrepancy creates a fundamental misalignment when patient-level supervision is directly imposed on conventional MIL frameworks, often leading to unstable optimization and degraded predictive reliability. To address this issue, we propose Anchor-Guided Evidence MIL (AGE-MIL), a weakly supervised framework for patient-level prediction. AGE-MIL constructs a patient-level anchor from slide representations to capture global pathological context and guide the retrieval and integration of diagnostically relevant local patches, enabling robust patient-level modeling. Patient-level risk is further modeled as an evidence accumulation process, promoting stable optimization under weak supervision. AGE-MIL is evaluated on six clinically relevant patient-level prediction tasks from two independent cohorts. Experimental results show that the proposed framework consistently outperforms eight state-of-the-art MIL methods. Code is available at https://github.com/wodeniua/AGE-MIL.

阅读与讨论 → 访问原文 →

14.

arXiv (CS.AI) 2026-06-12 DOI: arXiv:2606.13039

Fault Lines: Navigating Ethics and Responsible AI Where National Policy Meets Local Practice in Public Sector Transformation

作者:

Sitong Lyu↗Shabnam Taghiyeva↗Mohit Kukadia↗Denis Newman-Griffis↗

arXiv:2606.13039v1 Announce Type: cross Abstract: The UK government has adopted a pro-AI stance to help transform public service delivery in the face of severe financial pressures, but the path to translate this vision into responsible AI practice remains ill-defined. While UK policy is often set at the national level, local authorities are responsible for most public service delivery, and the rapid advance of AI-first narratives in the public sector is exposing fault lines in knowledge and practice at this national-local interface. This paper examines how responsible AI is interpreted and implemented at the interface between the UK's central government and local authorities, taking the high-stakes area of Special Educational Needs and Disabilities (SEND) as a case study. We present a thematic analysis of 17 semi-structured interviews with policymakers, practitioners, and third-sector professionals to identify barriers and enabling conditions for responsible AI where national policy meets local practice. We identify five interconnected challenges facing local authorities: shadow usage of AI and data privacy risks, market-government asymmetry in AI provision, insufficient workforce readiness, a lack of standardised definitions and measurements, and gaps in human accountability. For each, participants proposed actionable steps, from strengthening data protection frameworks and rebalancing the market-government relationship to enhancing workforce capacity. Our examination of SEND brings these challenges into sharper focus, showing how high-stakes decisions affecting vulnerable children and families intensify tensions around accountability, fairness, and human oversight, exposing the limits of a principle-based regulatory approach. We argue that responsible public sector AI requires both national policy adjustments and structural reforms to institutional capacity, values, and governance mechanisms at the local level.

阅读与讨论 → 访问原文 →

15.

arXiv (CS.LG) 2026-06-17 DOI: arXiv:2606.17192

Constrained Diffusion Models with Primal-Dual Inference

作者:

Samar Hadou↗Yigit Berkay Uslu↗Alejandro Ribeiro↗

arXiv:2606.17192v1 Announce Type: new Abstract: This paper develops constrained diffusion models with primal-dual inference (PDI) to sample from optimal distributions of entropy-regularized optimization problems with average constraints. We formalize constrained sampling in the Lagrangian dual domain, where the optimal distribution takes the form of a Gibbs distribution indexed by the optimal dual variable. Rather than estimating this dual multiplier before sampling and freezing it throughout generation, PDI jointly infers the optimal primal distribution and its parametrizing dual variable. Each reverse diffusion step denoises using the score field associated with the current multiplier and then updates the multiplier through dual ascent using the estimated constraint violation of the denoised samples. To enable this conditional score field, we train a single dual-conditioned score network over the family of Gibbs distributions induced by the dual variables encountered during inference. We prove that the time average of the dual variables generated along the inference trajectory converges to a neighborhood of the dual optimum and bound the effect of residual dual mismatch on the terminal distribution through schedule-dependent stability factors. We evaluate PDI on constrained sampling from a mixture of Gaussians, wireless resource allocation, and portfolio management.

阅读与讨论 → 访问原文 →

16.

medRxiv (Medicine) 2026-06-16 DOI: HASH:cb0c5506ec4c4440f9f7f7ee9c808ed8

Development of an automated, imaging-based preoperative screening model for early identification of malnutrition in an abdominal surgery cohort

作者:

Gershuni↗V. M↗Damani↗R. A↗Vasisht↗Sharma↗Rowe↗Compher↗Duda↗Sagreiya↗Kelz↗Lee↗…

Background: Clinical malnutrition affects one in five abdominal surgery patients and increases postoperative complications and mortality. Current screening occurs after admission, closing the window for preoperative nutritional intervention. No objective, scalable preoperative screening tool exists. Objective: To determine whether automated volumetric CT-based body composition analysis improves preoperative identification of surgical patients at risk for clinical malnutrition compared to clinical variables or single slice imaging alone. Methods: Retrospective cohort study of adults undergoing elective abdominal surgery at a quaternary academic medical center (2018 to 2021) with a preoperative CT scan within 90 days and complete nutrition assessment. Clinical malnutrition was diagnosed by a registered dietitian using ASPEN/AND criteria. Three sex stratified Elastic Net models were compared: (1) base clinical variables; (2) base plus L3 single slice skeletal muscle index and attenuation; and (3) base plus comprehensive 3D volumetric quantification of five muscle groups and two fat depots. Discrimination (AUROC), calibration (Brier score), and clinical utility (decision curve analysis) were assessed via 10-fold cross-validation. Results: Among 1,143 patients (52.4% female; mean age 60.5 years), 231 (20.2%) were diagnosed with malnutrition. Malnourished patients had significantly higher complication rates (36.4% vs. 15.4%, p

阅读与讨论 → 访问原文 →

17.

arXiv (CS.AI) 2026-06-16 DOI: arXiv:2606.14831

Is Your Agent Playing Dead? Deployed LLM Agents Exhibit Constraint-Evasive Fabrication and Thanatosis

作者:

Andoni Rodr\'iguez↗Alberto Pozanco↗Daniel Borrajo↗

arXiv:2606.14831v1 Announce Type: cross Abstract: This paper presents and characterizes a spectrum of previously unreported behaviours we term Constraint-Evasive Fabrication (CEF): when an LLM agent operates under irreconcilable constraints (where no response can simultaneously satisfy all active rules) it spontaneously fabricates plausible external obstacles and presents them as a fact. At the extreme end of this spectrum lies Constraint-Evasive Thanatosis (CET); the limit case where, rather than inventing a plausible excuse, the model simulates a full system crash to make the user disengage entirely. We first observed CET in an uncontrolled deployment test, where a GPT-4o banking agent fabricated Python-style exception traces (complete with memory addresses) to feign a system failure when threatened by a user. In subsequent controlled experiments, the model independently invented audit restrictions, microservice architectures, error codes, and service timeouts, none present in its prompt. Reproduction attempts across pressure levels and attacker personas yielded CEF consistently but with substantial variation in form, onset, and severity: the phenomenon is robust but stochastic. Critically, injecting ground-truth data mid-conversation did not restore honest behaviour once fabrication had taken hold (the model ignored correct information and continued confabulating) suggesting CEF is self-reinforcing rather than a knowledge gap. We show that (1) standard enterprise guardrails routinely create CEF-enabling conditions in production, (2) current RLHF procedures suppress but cannot eliminate CEF, and (3) existing safety benchmarks do not test for this failure mode. Our results highlight the need for irreconcilable-constraint benchmarks, CEF-aware training procedures, and deployment-time detection methods before constrained agents become further entrenched in high-stakes domains.

阅读与讨论 → 访问原文 →

18.

arXiv (CS.CL) 2026-06-19 DOI: arXiv:2606.20023

When Lower Privileges Suffice: Investigating Over-Privileged Tool Selection in LLM Agents

作者:

Kaiyue Yang↗Yuyan Bu↗Jingwei Yi↗Yuchi Wang↗Biyu Zhou↗Juntao Dai↗Songlin Hu↗Yaodong Yang↗

As LLM agents increasingly select tools autonomously, their choices among tools with different privileges become safety-relevant. However, prior tool-selection studies focus on safety-agnostic metadata preferences, leaving privilege-sensitive choices underexplored. To address this gap, we study over-privileged tool selection, in which an agent selects or escalates to a higher-privilege tool despite a sufficient lower-privilege alternative. We introduce ToolPrivBench to evaluate whether agents choose higher-privilege tools despite sufficient lower-privilege alternatives, measuring both initial selection and escalation after transient tool failures. Across eight domains and five recurring risk patterns, we find that over-privileged tool selection is common among mainstream LLM agents and is further amplified by transient failures. We further find that general safety alignment does not reliably transfer to least-privilege tool choice, while prompt-level controls provide only limited mitigation under transient failures. We therefore introduce a privilege-aware post-training defense that teaches agents to prefer sufficient lower-privilege tools and escalate only when necessary. Our mitigation experiments show that this defense substantially reduces unnecessary high-privilege tool use while preserving general capabilities.

阅读与讨论 → 访问原文 →

19.

arXiv (CS.AI) 2026-06-12 DOI: arXiv:2505.04021

Prism: Cost-Efficient Multi-LLM Serving via GPU Memory Ballooning

作者:

Shan Yu↗Yifan Qiao↗Mingyuan Ma↗Yangmin Li↗Shuo Yang↗Xinyuan Tong↗Yang Wang↗Zhiqiang Xie↗Yuwei An↗Shiyi Cao↗Ke Bao↗Deepak Vij↗…

arXiv:2505.04021v3 Announce Type: replace-cross Abstract: Inference providers must maintain availability for many LLMs, including low-volume but essential models, making resource efficiency increasingly important as token prices fall. Analysis of production traces reveals a dynamic bursty-group pattern in which sets of models become active together and shift over time; existing space- and time-sharing approaches lack principled mechanisms to adapt to this variability, forcing trade-offs between SLO adherence and efficiency. We observe that elastic memory allocation can unify spatial and temporal sharing. Based on this insight, we have developed Prism, a memory-centric LLM co-serving framework that applies memory ballooning to reclaim memory across models and support both forms of sharing under a single scheme. Prism's balloon driver, referred to as kvcached, has been open-sourced at https://github.com/ovg-project/kvcached, and deployed in production environments across 10K+ GPUs.

阅读与讨论 → 访问原文 →

20.

arXiv (CS.CV) 2026-06-17 DOI: arXiv:2606.17800

MaineCoon: Pursuing A Real-Time Audio-Visual Social World Model

作者:

Lichen Bai↗Tianhao Zhang↗Shitong Shao↗Dingwei Tan↗Qiyu Zhong↗Zhengpeng Xie↗Haopeng Li↗Qinghao Huang↗Dandan Shen↗Tengjiao Ji↗Wei Wang↗Peicheng Wu↗…

As an increasing majority of global video content is consumed on social platforms for interactive social purposes, video generation models built for social worlds are important but largely overlooked by previous studies. In this work, we define the position of social world models and build a prototype model as the first step towards this goal. While previous world models successfully simulate physical environments or gaming world exploration, they remain fundamentally detached from human-centric social dynamics. To bridge this gap as the first step to social world models, we present MaineCoon, the first real-time audio-visual autoregressive model that has 22B parameters and is capable of real-time streaming generation and sub-second interaction, with a record-breaking frame rate of up to 47.5 FPS, on a single GPU. To the best of our knowledge, MaineCoon is also the first real-time audio-visual generation model specifically optimized for social-interactive applications. To enable efficient and stable training, we introduce several novel techniques into MaineCoon, including self-resampling, cross-modal representation alignment, domain-aware preference optimization, and reinforced online-policy distillation (ROPD). We also design the first agentic streaming inference framework that supports thousand-second-scale or even longer generation while mitigating drift with agentic cache management and prompt planing. These innovations significantly accelerate training while optimizing real-time inference performance. We believe this work not only sets a new state-of-the-art (SOTA) performance benchmark for high-quality, low-latency, and long-horizon audio-visual autoregressive models, but also points out the paradigm shift desired for next-generation AI-native social platforms.

阅读与讨论 → 访问原文 →

21.

arXiv (CS.CV) 2026-06-12 DOI: arXiv:2602.00462

LatentLens: Revealing Highly Interpretable Visual Tokens in LLMs

作者:

Benno Krojer↗Shravan Nayak↗Oscar Ma\~nas↗Vaibhav Adlakha↗Desmond Elliott↗Siva Reddy↗Marius Mosbach↗

Transforming a large language model (LLM) into a vision-language model (VLM) can be achieved by mapping the visual tokens from a vision encoder into the embedding space of an LLM. Intriguingly, this mapping can be as simple as a shallow MLP transformation. To understand why LLMs can so readily process visual tokens, we need interpretability methods that reveal what is encoded in the visual token representations at every layer of LLM processing. In this work, we introduce LatentLens, a novel approach for mapping latent representations to descriptions in natural language. LatentLens encodes a large text corpus and stores contextualized token representations for each token in that corpus. Visual token representations are then compared to these contextualized representations and the top-nearest neighbor representations serve as descriptions of the visual token. We evaluate this method on 15 different VLMs, showing that commonly used methods, such as LogitLens, substantially underestimate the interpretability of visual tokens. With LatentLens instead, the majority of visual tokens are interpretable across all studied models and all layers. Qualitatively, we show that the descriptions produced by LatentLens are semantically meaningful and provide more fine-grained interpretations for humans compared to individual tokens. More broadly, our findings contribute new evidence on the alignment between vision and language representations and open up new directions for analyzing the latent representations of LLMs.

阅读与讨论 → 访问原文 →

22.

arXiv (quant-ph) 2026-06-16 DOI: arXiv:2603.23124

Quantum speedup from nonclassical polarization

作者:

Tim A{\ss}brock↗Jan Sperling↗Laura Ares↗

arXiv:2603.23124v2 Announce Type: replace Abstract: We develop a framework for identifying nonclassical speedups in systems with polarization, likewise spin degrees of freedom. By confining the dynamics to the manifold of angular momentum coherent states, which act as the classical reference in this case, we compute the speed limit that bounds the rate of change of the state achievable without generating quantum coherence. A comparison with the unrestricted quantum speed limit enables the quantitative identification of speedups arising from polarization nonclassicality. We apply this framework to the cross-Kerr interaction, demonstrating a persistent speedup scaling as $\mathcal{O}(\sqrt{N})$ with the photon number $N$ with a parity effect in favour of even photon numbers. The results establish polarization nonclassicality as a genuine dynamical resource, linking quantum coherence to quantum-enhanced evolution speeds in nonlinear photonic systems.

阅读与讨论 → 访问原文 →

23.

arXiv (CS.LG) 2026-06-11 DOI: arXiv:2510.01529

Bypassing Prompt Guards in Production with Controlled-Release Prompting

作者:

Jaiden Fairoze↗Sanjam Garg↗Keewoo Lee↗Mingyuan Wang↗

arXiv:2510.01529v4 Announce Type: replace Abstract: Ball et al. recently established that prompt filtering for AI alignment faces a fundamental barrier: under standard cryptographic assumptions, no filter running significantly faster than the protected model can universally distinguish adversarial prompts from benign ones. We investigate whether this impossibility result translates to real-world vulnerabilities in deployed large language model (LLM) systems. We answer affirmatively by introducing controlled-release prompting, a practical instantiation of the theoretical framework that exploits the resource asymmetry between lightweight input filters and the main models they protect. Unlike the theoretical construction, our attack does not require model modification: it generates malicious prompts that are indecipherable by any bounded filter yet remain tractable to the target LLM. We find our attack to be successful on four major chat platforms (Google Gemini, DeepSeek Chat, xAI Grok, and Mistral Le Chat) where baseline methods fail. Additionally, we apply our attack to extract copyrighted data from Gemini. Finally, we provide a systematic evaluation of 14 open-weight prompt guard models, revealing that even reasoning-capable filters cannot reliably detect our attack without incurring prohibitive resource overhead.

阅读与讨论 → 访问原文 →

24.

arXiv (CS.CV) 2026-06-17 DOI: arXiv:2512.09373

FUSER: Feed-Forward MUltiview 3D Registration Transformer and SE(3)$^N$ Diffusion Refinement

作者:

Haobo Jiang↗Jin Xie↗Jian Yang↗Liang Yu↗Jianmin Zheng↗

Registration of multiview point clouds conventionally relies on extensive pairwise matching to build a pose graph for global synchronization, which is computationally expensive and inherently ill-posed without holistic geometric constraints. This paper proposes FUSER, the first feed-forward multiview registration transformer that jointly processes all scans in a unified, compact latent space to directly predict global poses without any pairwise estimation. To maintain tractability, FUSER encodes each scan into low-resolution superpoint features via a sparse 3D CNN that preserves absolute translation cues, and performs efficient intra- and inter-scan reasoning through a Geometric Alternating Attention module. Particularly, we transfer 2D attention priors from off-the-shelf foundation models to enhance 3D feature interaction and geometric consistency. Building upon FUSER, we further introduce FUSER-DF, an SE(3)$^N$ diffusion refinement framework to correct FUSER's estimates via denoising in the joint SE(3)$^N$ space. FUSER acts as a surrogate multiview registration model to construct the denoiser, and a prior-conditioned SE(3)$^N$ variational lower bound is derived for denoising supervision. Extensive experiments on 3DMatch, ScanNet and ArkitScenes demonstrate that our approach achieves the superior registration accuracy and outstanding computational efficiency.

阅读与讨论 → 访问原文 →

25.

arXiv (quant-ph) 2026-06-16 DOI: arXiv:2606.15227

Cosmological Pseudo-Entropy

作者:

Manghang Limbu↗Pramod Kamal Kharel↗Rohit Bhattarai↗Kiran Adhikari↗

arXiv:2606.15227v1 Announce Type: cross Abstract: We study pseudo entropy $\mathcal{S}$, a recent generalization of entanglement entropy, for scalar cosmological perturbations in de Sitter space with sound speed $0.024 \leq c_s \leq 1$, and in expanding and contracting FLRW backgrounds with varying equation-of-state parameter $w$. In de Sitter space, $\mathrm{Re}(\mathcal{S})$ grows after horizon exit while $c_s$ controls its onset and saturates at late times. A similar saturation occurs in expanding-accelerating and contracting-decelerating backgrounds. In contrast, expanding-decelerating and contracting-accelerating backgrounds show large early-time $\mathrm{Re}(\mathcal{S})$ followed by oscillations after horizon re-entry. This happens because while the squeezing freezes, the squeezing angle doesn't. Unlike entanglement entropy, pseudo entropy possesses an imaginary part, $\mathrm{Im}(\mathcal{S})$, as well, which can encode the relative phase. $\mathrm{Im}(\mathcal{S})$ decays to zero in de Sitter and expanding-accelerating cases, but forms dense sub-Hubble oscillation bands in expanding-decelerating and contracting-accelerating backgrounds. Compared with entanglement entropy, Krylov complexity, and Nielsen circuit complexity, pseudo entropy captures otherwise hidden phase information; in the unsaturated regime, its slope is $\sqrt{2}$ times that of Nielsen complexity. Unlike circuit complexity, whose saturation bound is $w$-independent, pseudo entropy is sensitive to $w$ during the transition regime, making it a finer information theoretic diagnostic of cosmological dynamics.

阅读与讨论 → 访问原文 →