arXiv AI

arXiv:2610.28989v1 Announce Type: new Abstract: Customized agent deployments define policies over tools, resources, and execution conditions. Providing these policies in the task agent's prompt does not reliably prevent violations. Training a guard requires examples of tool calls that comply with or violate those policies. We introduce AgentPolicyForge, a framework that uses executable environments to construct policy benchmarks and iteratively train policy guards. With this framework, we construct PolicyGym, a dataset across ten policy families using 698 scenarios for training data generation and 254 separate held-out scenarios for human-reviewed test sets. Tool calls are executed and their outcomes recorded to help determine whether they comply with the policy. Every violating case uses a permitted tool, requiring checks of arguments, resources, and context rather than tool identity alone. The case construction process supports iterative guard training: an attacker generates requests using retrieved strategies, and the guard is fine-tuned on balanced, environment-verified examples. Training improves policy enforcement even with a fixed archive. Updating the archive further reduces in-domain attack success compared with keeping it fixed. We compare policy-aware task agent self-regulation with a trained policy guard. On PolicyGym, the final guard-based configuration achieves a 40-44% relative reduction in attack success rate and fewer benign rejections than policy prompting alone.

arXiv:2610.28987v1 Announce Type: new Abstract: Persistent memory is a largely undefended surface for injection and poisoning, and the same risk limits memory sharing among agents. We present PerMIT, a memory architecture with provable integrity guarantees under an explicit trust assumption: only user-typed inputs and values confirmed at a gate are trusted. PerMIT mediates the release and use of untrusted environment-derived values through a finite memory interface, while preserving agent-driven multi-turn planning and tool choice. Over this finite interface, we prove two integrity theorems that are also machine-checked in Lean 4, which together ensure that environment-derived memory reaches a sensitive action only if confirmed at the gate and influences the agent's behavior only through what the gate displays. Every attack in our threat model thereby reduces to a single auditable decision, which a human, an LLM verifier, or both can make in practice. On Trojan Hippo, PerMIT records zero observed attack success with self-verification and lightweight verifiers. Because each confirmation is a logged verdict on exact content, endorsements accumulate across users, and sufficiently endorsed items become community-confirmed with no further verification required. Under explicit assumptions, we prove and corroborate in simulation a co-scaling property: as the community grows, the risks of promoting poison and of wrongly evicting benign knowledge fall, while per-user verification effort does not rise. On $\tau^2$-bench, communities of up to 90 autonomous agents use PerMIT to build shared knowledge bases with no direct memory injection promoted, and the resulting knowledge significantly improves new agents on held-out tasks, suggesting a path toward agents that safely self-improve at test time.

arXiv:2610.28982v1 Announce Type: new Abstract: While large language models are increasingly deployed as tool-calling agents, their safety evaluations still focus primarily on their ability to refuse harmful requests in text. We investigate whether the refusal behavior in text transfers to tool use. We construct a benchmark of 2,304 prompts across four domains, consisting of matched triplets that present the same harmful or benign request in text-only, standard tool-enabled, and adversarially framed tool-enabled settings. We then assess tool-call safety using 20 predefined forbidden actions. Testing five models of varying sizes and families, we find that requests refused in text frequently result in unsafe tool calls. Under standard tool-enabled prompting, this divergence occurs for 10-52 percent of harmful requests refused in chat. To mechanistically investigate this gap, we identify a refusal-associated direction in the residual stream of each of the five models. We find that in tool-enabled contexts, the activations along this direction decrease in four of five models. Furthermore, lower activation predicts unsafe tool calls across all five models (AUC 0.76-0.86). However, restoring the signal to the respective text-only value through activation patching only prevents a subset of unsafe calls, primarily by suppressing tool use rather than by producing safer calls. Steering along the refusal direction nearly eliminates unsafe calls only at strengths that also suppress benign tool use or degrade the output quality. Our findings indicate that the identified refusal direction contributes to whether a model acts, but only partially explains the refusal divergence. Therefore, evaluating the safety of tool-calling agents requires directly examining their actions along with their textual responses.

arXiv:2610.28945v1 Announce Type: new Abstract: Large language model (LLM) agents are increasingly expected to interact with environments and improve their decisions during deployment. However, how anticipating future outcomes and drawing on accumulated interaction experience can jointly support decision-making remains underexplored. We propose a test-time framework, inspired by Model Predictive Path Integral (MPPI) control, that combines multi-branch predictive analysis with experience-based adaptation. At each step, the agent proposes multiple candidate actions and imagines their outcomes with a world model, so that alternatives can be compared without being executed in the environment. Since language environments provide neither explicit dynamics nor costs, the predicted outcomes are evaluated for task progress and risk, and MPPI-style weights over the resulting costs guide the final decision. A dual-branch experience memory enables self-improvement: the planner branch refines candidate generation and action selection, while the world model branch improves next-observation prediction. Experiments on ALFWorld and ScienceWorld across three backbones demonstrate the strong overall performance of our method and the effectiveness of experience-guided predictive control. Our code will be made publicly available.

arXiv:2610.28934v1 Announce Type: new Abstract: Coding agents now complete tasks autonomously, modifying real-world state such as artifacts, software environments, and configurations along the way. Yet they show limited backup awareness: recognizing that an action will modify state that may later be needed, and preserving it beforehand. To evaluate backup awareness, we introduce RegretBench, a benchmark of 205 executable tasks grounded in regret moments reported on platforms such as X.com, GitHub, and Stack Overflow, where users explicitly regretted not preserving state they later needed. Across six frontier models, agents complete 86.1% of tasks on average yet back up before these moments in only 15.2%, showing that task completion ability does not imply backup awareness. Because it is hard to predict which changes will be regretted, we propose BackUpper, a harness plug-in backup layer that checks every tool call and preserves affected state before modification. It combines rule-based strategies for common operations with a lightweight subagent for complex cases, and its rule table can expand with subagent decisions over time. On RegretBench, BackUpper covers every regret moment action and restores states with 92.2% average similarity to the original. Achieving this costs only $0.014 per task, with 7x fewer tokens than the main-agent self-backup baseline. When users have preferences on what to back up, BackUpper can also narrow its scope accordingly, matching human annotations on whether an action is worth preserving with an 84.5% F1 score. These results suggest that agents can keep actions reversible at low cost while respecting users' preferences.

arXiv:2610.28914v1 Announce Type: new Abstract: Frontier language-model agents excel at complex tasks, but high inference costs make transferring their capabilities to smaller models essential. Traditional knowledge distillation achieves this through parameter updates, requiring access to the student's weights and additional training compute while encoding the transferred knowledge implicitly in the parameters. We introduce D-Skill, a framework that treats the teacher as a black box and distills its behavioral advantage into an interpretable natural-language skill by analyzing paired rollouts and identifying behavioral divergences and agreements. The resulting lessons are consolidated offline, with any compatible skill-update backend, into a skill document that is provided to the student at inference time. D-Skill constructs the skill entirely offline from fixed trajectories, without labels, further environment interaction, or weight updates after rollout collection. Across SpreadsheetBench-Verified, ALFWorld, and chess puzzles spanning multiple teacher-student pairs, D-Skill closes 75%-95% of the teacher-student gap without touching model weights. After a single offline pass over the paired traces, D-Skill outperforms one-epoch SFT baselines and remains competitive with multi-epoch SFT. Under the same model family, the resulting student also incurs 5.7x-8.3x lower inference cost per task than the teacher. Our results demonstrate that natural-language skill distillation offers a performant, interpretable, and efficient alternative to weight-based capability transfer.

arXiv:2610.28912v1 Announce Type: new Abstract: This paper charts the evolution of the Prolog textbook Simply Logical -- Intelligent Reasoning by Example from print with runnable programmes on a 3.5-inch diskette published in 1994, via various intermediate online versions, to the fully interactive online edition available today. Our aim is to investigate -- from both the writer's and the reader's perspectives -- the potential impact of technology on educational materials. The three authors of this paper present three distinct and complementary points of view, which come together to shape an interactive online book that offers an immersive learning experience. Peter describes the philosophy behind the original book and experiences teaching from it. Jan demonstrates how contemporary web programming has enabled a fully interactive realisation of the book's philosophy. Kacper reports how different technologies can be adopted to build a versatile authoring toolkit that underpins the interactive online edition of the book. Collating these distinct yet coherent perspectives allows us to reflect on future opportunities for interactive Prolog teaching materials arising from the continuous development of web technologies. The insights we offer should be of interest to teachers of Prolog as well as those wanting to find out about the latest educational technologies.

arXiv:2610.28909v1 Announce Type: new Abstract: Deep research systems iteratively plan, gather evidence, synthesize, and revise, making end-to-end latency a key bottleneck for interactive use. Almost all production systems are built on autoregressive (AR) language models and inherit left-to-right decoding in planning and long-form writing. We present DiffResearch, an open multi-agent deep research framework that places a diffusion language model (dLLM) at the core of every stage, and measure how a diffusion backbone changes latency and report quality. Holding the agentic scaffold, retrieval code, prompts, and evaluation fixed, we swap the backbone between an open diffusion model and a comparable AR model. Across 100 Deep Research Bench tasks, the diffusion backbone is roughly 6 times faster end-to-end, driven by order-of-magnitude reductions in planning and writing. We further introduce dFast, a decomposition mode that reads intermediate denoising states and extracts subqueries before generation finishes, providing a 1.99 times median speedup for decomposition in isolation with no loss of report quality. On the benchmark's RACE rubric, dFast obtains the best overall score among the three configurations. Finally, in a public system-level leaderboard snapshot, an instance built on a stronger diffusion model ranks above several commercial systems. We release DiffResearch under an Apache-2.0 license.

arXiv:2610.28905v1 Announce Type: new Abstract: Diffusion language models form answers through a sequence of revisable continuous states, but multi-agent systems usually discard those states and make agents communicate only after decoding text. We ask whether one independent diffusion trajectory can pass a compact hidden message into another and, crucially, whether the receiver uses what that message means. In Hidden Messages, a sender compresses its denoising state into four signed dense vectors; the receiver consumes them while retaining its own context, canvas, and trajectory. Removing a message tests whether the channel matters; replacing it with a message from another example tests whether its content matters. Across controlled tasks, derangement removes a +63.6-point advantage on Tiny-A2D and a +100.0-point advantage on Dream-7B, while targeted substitutions steer the receiver toward the substituted fact. On QASC with the two supporting facts split across agents, matched communication reaches 80.8%, but a wrong-example message lowers accuracy to 46.8%, a content effect of +34.0 points. The surprising result is temporal: one late transfer reaches 87.8%, exceeding four recurrent exchanges by +7.0 points. These experiments provide causal evidence for continuous hidden communication between diffusion-LM agents and show that when information is transferred can matter more than how many times agents exchange it.

arXiv:2610.28891v1 Announce Type: new Abstract: Many social evaluations for multiagent systems study behaviors as they emerge from environments instead of systematically intervening in the environment to elicit such behaviors. We introduce Fig (Finding Interactions in Games), an agentic system that performs targeted elicitation interventions on game environments. It takes a natural language behavior description, decomposes it into sub-behaviors, generates new social games or selects existing ones, and executes single or combined interventions to the games' engine, opponents, instructions, and context. To evaluate model behavior, it combines built-in mechanical checks and synthetic rubric judgments. We validate the interventions, generated games, and evaluation mechanisms via human review. Models attempt the target behavior in 45.2% of episodes on average for valid scenarios generated by Fig for six test behaviors -- cooperation, exploitation, deception, covert coordination, betrayal, and sabotage. For game-behavior pairs with available controls, Fig's interventions increase attempt rates by 10.9% on average compared to matched control games. Fig shows that multiagent environments can serve as programmable test beds for intentional, controllable, and reproducible behavior evaluation. Code is available at https://github.com/fleet-ai/fig.

arXiv:2610.28875v1 Announce Type: new Abstract: Large language models increasingly take part in extended conversations where they can both shape and be shaped by another party's views, yet most persuasion research measures only one side of that exchange. We introduce a controlled LLM-LLM debate framework in which two language-model roles repeatedly discuss a contested topic (climate-change attribution): one is given an assigned stance and, in some conditions, an explicit instruction to persuade, while the other receives no assigned position. Across four ordered model combinations involving GPT-6 Luna and Gemini 2.5 Flash Lite, we track both each role's explicitly reported position and a separate measure of its turn-by-turn behavior inside the conversation. Opinion dynamics differ substantially across model combinations: same-model interactions narrow the opinion gap on average, clearly for only one of the two models, while cross-model interactions widen it. Explicit reports and in-dialogue behavior capture distinct aspects of opinion dynamics, and a blanket instruction to persuade does not reliably increase influence. These findings suggest that conversational influence depends not only on assigned roles and initial positions but also on the participating models, with implications for how opinion change is measured in extended LLM interactions.

arXiv:2610.28855v1 Announce Type: new Abstract: Progress on foundation model pretraining is gated by who can afford a trial, and relational pretraining is no exception. Recipes are developed inside individual labs, reported under private protocols, and do not accumulate. Speedrun leaderboards remove that barrier by decentralizing the search rather than the training run. The problem is fixed, the stack is open, entries are ranked by wall-clock time to a quality target, and one GPU is enough to hold the record. We introduce modded-nanoRT, the first such leaderboard for Relational Transformers. Which database to hold out decides whether the benchmark can resolve improvements at all, so we fix it by calibration, ranking six candidates by how sharply a 45-minute learning curve separates improvement from seed noise. We open the leaderboard with typed trilinear attention, a relational operator that conditions attention scores and values on the foreign key traversed, together with Muon and SIGReg. The recipe reaches the target in 305.63 seconds on one NVIDIA H200, 8.11x faster than the reference implementation, and wins 13 of 15 pairs outside the development database. We will release the benchmark, its frozen artifacts, and the record runs. Our code is available at https://github.com/kunumi/nanoRT.

arXiv:2610.28836v1 Announce Type: new Abstract: This study proposes a sketch-to-simulation system to solve handwritten circuit diagrams. The system converts raw images into a Simulation Program with Integrated Circuit Emphasis (SPICE)-compatible netlist and automatically calculates the circuit's nodal voltages and currents. Another objective is to propose a holistic framework that benefits students and educators by enhancing the teaching and learning experience of electric circuits and expediting engineering design reviews. This article presents a multistage pipeline that converts raw photographs of handwritten electric circuits into executable SPICE-like netlists, providing a solution. A dataset of 1,000 hand-drawn Resistor-Inductor-Capacitor (RLC) circuits was curated and annotated for components, text regions, and character recognition. The proposed system decomposes a circuit diagram into three stages: (i) dual You Only Look Once (YOLO)-based detectors for components and text regions, (ii) a Convolutional Recurrent Neural Network (CRNN) with Connectionist Temporal Classification (CTC) and Optical Character Recognition (OCR)-based domain-aware symbol value extraction, and (iii) image processing with thresholding, skeletonization, and Density-Based Spatial Clustering of Applications with Noise (DBSCAN) to automatically identify nodes and map them with corresponding components. The proposed system produces the netlist and solves the circuit. The qualitative and quantitative ablation studies on modular stages, as well as end-to-end tests, demonstrate robust performance across various handwriting styles, lighting conditions, and clutter.This study serves as an interactive learning aid and practical tool for professionals, providing a comprehensive user interface (UI), enabling rapid digitization of circuit sketches for study, design, prototyping, or troubleshooting.

arXiv:2610.28826v1 Announce Type: new Abstract: Automated judges of clinical artificial intelligence (AI) outputs are necessary, but large language model (LLM) judges are costly and slow at scale. We compared Jev, a decision language model that returns structured judgments, with frontier and open-weight LLM judges in four use cases. On 118 pediatric answers, Jev's agreement with physicians' ordinal quality ratings (Gwet's AC2 0.50; expected score 0.53) matched or exceeded that of seven of 12 LLM judges, including frontier models, Claude Opus 5.5 (0.49) and GPT-6 (0.38). On physician-graded binary HealthBench criteria, Jev matched or exceeded four of six LLM judges. On MedHELM's open-ended tasks, Jev's agreement with a three-LLM jury approached or exceeded the jurors' internal agreement. On HealthBench Professional, Jev matched the benchmark's reference judge, GPT-5.4, on 91% of criteria (other judges, 91-93%). Frontier judges cost 73-294 times more than Jev and took 7-57 times longer. Jev may enable efficient clinical AI evaluation at scale.

arXiv:2610.28815v1 Announce Type: new Abstract: Agentic harness evolution optimizes an agent's instructions, tools, and control flow by using its task execution experiences as evolving evidence. While more evolution tasks should intuitively improve performance, our preliminary study reveals a paradoxical failure mode: the performance of existing methods degrades as the task pool expands. Our further analysis demonstrates that this degradation stems from suboptimal evidence selection during optimization, as different evidence combinations lead to substantially different evolution outcomes. To address this issue, we propose Successive Scale Racing (SSR), a framework that adaptively searches for optimal evidence combinations in each evolution round. SSR constructs diverse evidence combinations based on the corresponding task quality and coverage, and efficiently races the resulting candidate harnesses under a small compute budget. Empirically, across three distinct benchmarks, SSR achieves state-of-the-art results in all six configurations. Notably, as the task pool grows from 40 to 200, SSR steadily improves the pass rate from 64.3% to 68.3%, while the baselines' performances all drop. Furthermore, its racing mechanism identifies near-optimal candidates while reducing validation executions by 60%.

arXiv:2610.28763v1 Announce Type: new Abstract: Frontier coding agents are increasingly framed as end-to-end automated researchers, and ML for science offers a demanding test of that promise: improving a scientific model often requires domain insight beyond routine ML engineering. We introduce Agent4Sci, a long-horizon testbed for measuring how far today's agents can improve the published best models on scientific ML benchmarks, spanning 20 benchmarks across 11 domains. Rather than ranking agents on a leaderboard, Agent4Sci asks what kinds of progress agents can make on these tasks under in-the-wild conditions, and where they fall short. Empirically, today's frontier agents advance the published best on every benchmark, with Claude Opus reaching 10.4% mean relative improvement and no run flagged for academic misconduct. These gains, however, come almost entirely from a bag of ML-engineering tricks rather than research-level insight. Trajectory analysis surfaces three recurring limitations in this setting: novelty miscalibration, resource under-utilization, and rapid collapse into local engineering sweeps. Lightweight plug-in skills for research ideation, experiment management, and scheduled checklists raise Claude Opus from 10.4% to 20.1% and Codex from 4.4% to 8.1%, while yielding more diverse, domain-relevant interventions. Agent4Sci gives the community a way to track and push frontier agents from ML engineering toward domain insight on real-world ML-for-science problems. Code and data are available on GitHub (https://github.com/maszhongming/agent4sci) and Hugging Face (https://huggingface.co/datasets/MingZhong/agent4sci).

arXiv:2610.28706v1 Announce Type: new Abstract: LLM-driven evolutionary search is a promising paradigm for scientific discovery, but existing methods remain costly and slow, often applying expensive reasoning uniformly while underusing evidence from failed, redundant, and saturated trials. We introduce Speculative Evolution (SpecEvo), a multi-LLM framework that coordinates lightweight and frontier models through continual consolidation of search evidence. Lightweight models power both parallel Speculators, which generate and refine candidate programs, and an Advisor, which distills accumulated successes, failures, and saturated directions into persistent guidance for subsequent exploration. At sparse checkpoints, a frontier-model Navigator inspects the evolving population and intervenes through synthesis, surgical refinement, or reframing according to search stagnation. This speculate-then-consolidate workflow delegates routine exploration and evidence consolidation to inexpensive models while reserving frontier reasoning for strategic interventions whose benefits extend across many subsequent low-cost evaluations. Across 176 tasks spanning four scientific and algorithmic discovery domains, SpecEvo matches or outperforms frontier-model baselines while reducing API cost by up to 80%, achieves up to 1.5x greater evaluation efficiency than cost-efficient baselines on equation discovery, and improves over the strongest CO-Bench baseline by 3.8%. These results support trajectory-conditioned model coordination as an effective approach to cost-efficient scientific discovery.

arXiv:2610.28698v1 Announce Type: new Abstract: Scaling data, diversity and model capacity has improved embodied learning in physical environments, where diverse experiences remain grounded in shared physical laws. Whether these gains extend to virtual worlds with different visual styles, dynamics, controls, and objectives remains unclear. We investigate this question through a controlled study of continued pre-training (CPT) across three axes: data volume, diversity, and model size. Our 120 training configurations span four model sizes from 0.8B to 27B, six data budgets from 50 to 1,600 hours, and five mixtures ranging from Minecraft-only data to diverse multi-world corpora. We assess performance and generalization using three-level validation losses and gameplay benchmarks from in-corpus worlds to unseen worlds. Our results show: (1) more data improves in-corpus performance, diversity improves transfer to related worlds with limited gains in distant worlds, and model size improves data efficiency but generalizes worse to distant worlds under single-world training; (2) we observe an early interface phase with broad cross-world benefits and a later world phase with gains narrowed to in-corpus and related worlds, and fit an empirical scaling law relating validation loss to data volume, diversity and model size; and (3) our models are competitive with Minecraft specialists and frontier models in 3D worlds, and perform best in real-time settings. However, the limits of imitation and persistent gaps in distant worlds suggest that building generalist embodied agents requires richer training stages beyond continued pre-training.

arXiv:2610.28686v1 Announce Type: new Abstract: Query clarification is essential for retrieval-augmented generation (RAG) systems that face ambiguous user queries. We consider multi-turn grounded conversational scenarios where a user query is either ill-formed or ambiguous in the context of the prior dialog or, more interestingly, it may be well-formed but when posed against retrieved documents there are multiple relevant interpretations. The task is to recognize both such cases and then generate an appropriate clarification request to the user. We present ClariEdge, a new dataset of approximately 3,900 examples that include such ambiguous queries along with negative examples that do not require clarification. We also show that a lightweight fine-tuned LoRA achieves 96.6% classification accuracy and outperforms prompted baselines of many larger models on both classification and quality of clarification generation. Our study shows that hard negatives are an essential ingredient to separate out queries that are truly ambiguous from queries that may look ambiguous but are not (i.e., can be resolved from the context). The same LoRA achieves high performance in both pre-retrieval and post-retrieval scenarios; thus, it can be used in two places in a RAG pipeline as a check before answer generation. The ClariEdge dataset is available at https://huggingface.co/datasets/ibm-research/ClariEdge.

arXiv:2610.16979v3 Announce Type: replace-cross Abstract: Generative retrieval (GR) offers a promising paradigm for industrial search by unifying candidate generation within an autoregressive framework. However, effectively leveraging user behaviors in industrial GR remains challenging: existing SID tokenization methods struggle to preserve collaborative signals while maintaining balanced codebook utilization, whereas reinforcement learning (RL) for prioritizing high-value candidates suffers from sparse rewards when generated SIDs fail to match positive items. To address these challenges, we propose WeGR, a unified GR framework with two complementary components. Specifically, the Collaborative and Collapse-free Tokenizer (CoCo-VAE) injects query-to-item and item-to-item collaborative signals into continuous representations before quantization, preserving behavioral information without disrupting codebook optimization. It further couples first-layer codebook refresh to reactivate underutilized codewords with fidelity anti-collapse regularization, which broadens gradient coverage over relevant codewords and penalizes collisions in deeper layers to yield behavior-aware SIDs with balanced utilization. For preference alignment, Dynamic Horizon Policy Optimization (DHPO) progressively expands reward evaluation and policy optimization horizons to alleviate reward sparsity, enabling coarse-to-fine preference learning from SID prefixes to complete items without auxiliary reward models. Extensive offline experiments validate that WeGR improves retrieval and ranking performance. Moreover, a six-week online A/B test on a large-scale e-commerce search platform achieves relative gains of 2.539% in orders, 1.508% in CTR, and 1.381% in CVR, while reducing CPU and GPU usage by 25.5% and 26.1%. These results highlight the effectiveness and efficiency of WeGR in industrial deployment, supporting multiple search scenarios and tens of millions of daily page views.