ICML 2026 Papers

Progress in neural combinatorial optimization for Dynamic Flexible Job Shop Scheduling Problem (DFJSP) is currently hindered by a methodological tension: static benchmarks encourage benchmark overfitting, while uncalibrated generators obscure algorithmic capability with stochastic noise. To resolve this, we introduce \textbf{DynaSchedBench}, a diagnostic framework for DFJSP that rigorously controls the instance-generation process. Instead of relying on parameter sampling, our approach utilizes Sequential Event-Space Calibrator (SESC) that computes a novel Schedule Stress Index (SSI) to stratify instances by difficulty. We demonstrate that SESC is substantially more computationally efficient than evolutionary baselines while converging reliably to the target metrics. The framework integrates modular components for instance generation, snapshot-based simulation, agents, evaluation, and visualization, thereby enabling rigorous testing of reactive and lookahead-based policies. Leveraging this calibrated environment, we identify key limitations of LLM-based scheduling agents. Specifically, in step-wise online decision-making for dynamic scheduling, we identify an ``Observability Paradox'': providing agents with oracle access to full structural information can degrade policy performance, underperforming concise information. Furthermore, despite substantial token overhead, tool-augmented and refinement strategies fail to reliably improve performance, and most LLM agents fail to consistently surpass strong dispatching baselines—behaving more like robust heuristic approximators than superior optimizers.

A number of scientific conferences and journals have recently enacted policies that prohibit LLM usage by peer reviewers, except for polishing, paraphrasing, and grammar correction of otherwise human-written reviews. But, are these policies enforceable? To answer this question, we assemble a dataset of peer reviews simulating multiple levels of human-AI collaboration, and evaluate five state-of-the-art detectors, including two commercial systems. Our analysis shows that all detectors misclassify a non-trivial fraction of LLM-polished reviews as AI-generated, thereby risking false accusations of academic misconduct. We further investigate whether peer-review-specific signals, including access to the paper manuscript and the constrained domain of scientific writing, can be leveraged to improve detection. While incorporating such signals yields measurable gains in some settings, we identify limitations in each approach and find that none meets the accuracy standards required for identifying AI use in peer reviews. Importantly, our results suggest that recent public estimates of AI use in peer reviews through the use of current AI-text detectors should be interpreted with caution, as they misclassify mixed reviews (collaborative human-AI outputs) as fully AI generated, potentially overstating the extent of policy violations.

Designing microstructures with coupled cross-physics objectives is a fundamental challenge where traditional topology optimization is often computationally prohibitive and deep generative models frequently suffer from physical hallucinations. We introduce AutoMS, a multi-agent neuro-symbolic framework that reformulates inverse design as an LLM-driven evolutionary search. AutoMS leverages LLMs as semantic navigators to decompose complex requirements and coordinate agent workflows, while a novel Simulation-Aware Evolutionary Search (SAES) mechanism handles low-level numerical optimization via local gradient approximation and directed parameter updates. This architecture achieves a state-of-the-art 83.8\% success rate on 17 diverse cross-physics tasks, significantly outperforming both traditional evolutionary algorithms and existing agentic baselines. By decoupling open-ended semantic orchestration from simulation-grounded numerical search, AutoMS provides a robust pathway for navigating complex physical landscapes that remain intractable for standard generative or purely linguistic approaches.

Large language models (LLMs) exhibit strengths across diverse domains. However, achieving strong performance across these domains with a single general-purpose model typically requires scaling to sizes that are prohibitively expensive to train and deploy. On the other hand, while smaller domain-specialized models are much more efficient, they struggle to generalize beyond their training distributions. To address this dilemma, we propose FusionRoute, a robust and effective token-level multi-LLM collaboration framework in which a lightweight router simultaneously (i) selects the most suitable expert at each decoding step and (ii) contributes a complementary logit that refines or corrects the selected expert’s next-token distribution via logit addition. Unlike existing token-level collaboration methods that rely solely on fixed expert outputs, we provide a theoretical analysis showing that pure expert-only routing is fundamentally limited: unless strong global coverage assumptions hold, it cannot in general realize the optimal decoding policy. By augmenting expert selection with a trainable complementary generator, FusionRoute expands the effective policy class and enables recovery of optimal value functions under mild conditions. Empirically, across both Llama-3 and Gemma-2 families and diverse benchmarks spanning mathematical reasoning, code generation, and instruction following, FusionRoute outperforms both sequence- and token-level collaboration, model merging, and direct fine-tuning, while remaining competitive with domain experts on their respective tasks.

To navigate partially observable visual environments, recent VLM agents increasingly internalize world modeling capabilities directly into their policies via explicit CoT reasoning with reinforcement learning (RL). However, mere passive exploitation of reasoning on visited states is insufficient for sparse-reward agentic tasks, as it lacks the epistemic drive to actively uncover the *known unknown* required for robust generalization. We ask: *Can VLM agents actively find signals that challenge and update their internal world model through curiosity-driven exploration?* In this work, we propose **GLANCE**, a unified framework that bridges reasoning and exploration by grounding the agent's linguistic world model into the stable visual representations of an evolving target network. Crucially, **GLANCE** leverages the discrepancy between linguistic prediction and visual reality as an intrinsic curiosity signal within reinforcement learning, steering the agent to actively explore areas where its internal model is uncertain. Extensive experiments across a series of agentic tasks show the effectiveness of **GLANCE**, and demonstrate that aligning *what the agent thinks* with *what the agent sees* is key to solving complex or sparse agentic tasks.

An important emerging application of coding agents is *agent harness optimization*: the iterative improvement of a *target agent* by editing and evaluating its code. Despite its relevance, the community lacks a systematic understanding of coding agent performance on this task. Harness optimization differs from conventional software engineering: agent harnesses interleave deterministic code with stochastic LLM completions, requiring structured capture of both intermediate execution traces and downstream outcomes. To address these challenges, we introduce (1) VeRO (Versioning, Rewards, and Observations), an *outer harness* that provides versioned snapshots, budget-controlled evaluation, and structured execution traces of *target harnesses*, and (2) VeRO-Bench, a benchmark suite of target agents and tasks with reference evaluation procedures. Using VeRO, we conduct an empirical study comparing optimizers across tasks and analyzing which modifications reliably improve target agent harnesses. We release VeRO to support research on agent optimization as a core capability for coding agents. Code is available at https://github.com/scaleapi/vero.

Tool-augmented LLM agents can harbor implicit state that persists across sessions, activates through events, and propagates across agent boundaries—all invisible to standard debugging. We formalize this as Persistent Semantic Entities (PSE): constructs defined by name binding, event triggering, and cross-boundary propagation, and evaluate them across 20 models from 9 families (3B–671B parameters). First, all tested models are susceptible to contamination (20–100%), with name binding as the dominant mechanism: without it, contamination is 0%. Second, persistence depends on contamination type rather than scale or deployment—factual injection conflicting with parametric knowledge self-corrects, but preference, persona, and instruction contamination persists at 100% over a 10-turn horizon with no decay, an effect that holds across providers in our controlled setting and is consistent with a model-intrinsic explanation. Third, context-isolated self-verification achieves 20–79% reduction without oracle references while keyword-based detection produces systematic false positives, and contamination compounds 3× across multi-agent pipelines. Preference and instruction contamination—persistent, lacking self-correction, and invisible to standard monitoring—represents a particularly concerning attack surface for deployed agent systems.

Group-based reinforcement learning (RL) methods have achieved remarkable success in improving the performance of large language models (LLMs) and have been rapidly extended to agentic tasks. However, their credit assignment relies heavily on coarse-grained trajectory-level attribution according to final outcomes, making it difficult to capture the contribution of individual steps, such as valuable steps obscured within failed trajectories. To uncover latent information and enable more faithful step-level credit assignment, we propose Graph-based Group Policy Optimization (GraphGPO), which first aggregates all rollout trajectories into a unified state-transition graph and then estimates the distance from each state to the task goal using the global information encoded in the graph. Finally, GraphGPO assigns credit to each edge by estimating a graph-based advantage, based on how much the transition reduces the distance to the task goal. In this way, GraphGPO significantly improves training efficiency and achieves state-of-the-art performance across a range of challenging benchmarks.

Open-domain multimodal document retrieval aims to retrieve specific components (paragraphs, tables, or images) from large and interconnected document corpora. Existing graph-based retrieval approaches typically rely on a uniform similarity metric that overlooks hop-specific semantics, and their rigid pre-defined plans hinder dynamic error correction. These limitations suggest that a retriever should adapt its reasoning to the evolving context and recover intelligently from dead ends. To address these needs, we propose **Failure is Feedback**, which casts subgraph retrieval as a *sequential decision process* and introduces two key innovations. (i) We introduce a *history-aware backtracking mechanism*; unlike standard backtracking that simply reverts the state, our approach piggybacks on the context of failed traversals, leveraging insights from previous failures. (ii) We implement an *economically-rational agentic workflow*. Unlike conventional agents with static strategies, our orchestrator employs a cost-aware traversal method to dynamically manage the trade-off between retrieval accuracy and inference costs, escalating to intensive LLM-based reasoning only when the prior failure justifies the additional computational investment. The project page is available at [failureisfeedback.github.io](https://failureisfeedback.github.io/).

Large language model (LLM) agents increasingly coordinate in multi-agent systems, yet we lack an understanding of where and why cooperation fails. Many real-world coordination problems are not social dilemmas: helping others—sharing documentation, unblocking a teammate—costs the helper almost nothing while producing substantial collective benefit. Whether LLM agents cooperate in this regime, where helping is free and they are explicitly instructed to do so, remains unknown. We build a turn-based multi-agent environment that strips away all strategic complexity, making cooperation costless and trivially optimal. Across eight widely used LLMs, capability does not predict cooperation: OpenAI o3 reaches only 17\% of optimal collective performance while the weaker o3-mini reaches 50\%, despite identical instructions to maximize group revenue. Using a causal decomposition that automates one side of agent communication, we separate cooperation failures from competence failures, and find that several capable models actively withhold information despite gaining nothing from withholding. Targeted interventions address each mode: explicit protocols roughly double the performance of competence-limited models, while small sharing incentives unlock cooperation-limited ones. Our results suggest that scaling intelligence alone will not solve coordination in multi-agent systems, and will require deliberate cooperative design, even when helping costs nothing.

Large language models (LLMs) are rapidly acquiring capabilities relevant to biological research, from literature synthesis to interpretation of experimental data. Increasingly, LLM agents can also perform *in silico* biology tasks that previously required experienced human biologists. These emerging AI capabilities offer new opportunities for scientific discovery and biomedical advances, but they also shift the landscape of biosecurity risks. To address this, we introduce the Agentic Bio-Capabilities Benchmark (ABC-Bench), a suite of tasks to measure agentic biosecurity-relevant capabilities. ABC-Bench evaluates LLM agents on both benign and dual-use biology tasks: writing code to operate liquid handling robots, designing DNA fragments for in vitro assembly, and evading DNA synthesis screening. These tasks require a combination of biology and software expertise. All tested LLM agents outperformed the median expert human baseliner on all three tasks. Agents performed highly on tasks drawing on published knowledge and well-documented protocols, and more weakly on a task requiring novel bioinformatics reasoning. In three wet-lab validation experiments, we found that OpenAI’s o4-mini-high produced scripts that, when run on an OpenTrons liquid handling robot, successfully assembled DNA with expected sequences.

Multi-agent debate has shown promise for improving the reasoning of large language models, yet recent theory suggests its benefits are highly regime-dependent. While interaction can amplify informative signals under corrective conditions, symmetric debate dynamics are neutral in expectation, often making majority voting preferable. We reconcile these views by arguing that debate is effective only when invoked at the right time and with appropriate structure. Based on this insight, we propose LASE: Leader-Adaptive Structured Engagement, a leader-centric multi-agent debate framework that selectively engages interaction only in non-neutral regimes. LASE introduces an asymmetric leader–supporter architecture that enables directed information flow and selective signal amplification, while defaulting to simple aggregation otherwise. Experiments across diverse reasoning benchmarks show that LASE achieves multi-agent-level performance with near single-agent token cost, substantially improving efficiency over static debate and voting baselines.

Autonomous web agents are increasingly deployed for long-horizon tasks, yet their ability to adhere to real-world policies remains critically underexplored compared to standard safety objectives. To address this gap, we introduce PolicyGuardBench, a benchmark of 60k policy-trajectory pairs designed to evaluate compliance through both full-trajectory and novel prefix-based violation detection tasks. Using this dataset, we train PolicyGuard, a lightweight guardrail model that achieves strong detection accuracy while maintaining high inference efficiency. Notably, our model demonstrates robust generalization capabilities, preserving high performance even on unseen domains. These contributions establish a comprehensive framework for studying policy compliance, showing that accurate and generalizable guardrails are feasible at small scales.

Automated design of multi-agent interactions with desirable equilibrium outcomes is inherently difficult due to the computational hardness, non-uniqueness, and instability of the resulting equilibria. In this work, we propose the use of game-agnostic _differentiable equilibrium blocks_ (DEBs) as modules in a novel, differentiable framework to address a wide variety of incentive design problems from economics and computer science. We call this framework _deep incentive design_ (DID). To validate our approach, we examine three diverse, challenging incentive design tasks: contract design, machine scheduling, and inverse equilibrium problems. For each task, we train a single neural network using a unified pipeline and DEB. This architecture solves the _full distribution_ of problem instances, parameterized by a context, handling _all_ games across a wide range of scales (from two to sixteen actions per player).

Assisting non-expert users to develop complex interactive websites has become a popular task for LLM-powered code agents. However, existing code agents tend to only generate frontend web pages, masking the lack of real full-stack data processing and storage with fancy visual effects. Notably, constructing production-level full-stack web applications is far more challenging than only generating frontend web pages, demanding careful control of data flow, comprehensive understanding of constantly updating packages and dependencies, and accurate localization of obscure bugs in the codebase. To address these difficulties, we introduce FullStack-Agent, a unified agent system for full-stack agentic coding that consists of three parts: (1) FullStack-Dev, a multi-agent framework with strong planning, code editing, codebase navigation, and bug localization abilities. (2) FullStack-Learn, an innovative data-scaling and self-improving method that back-translates crawled and synthesized website repositories to improve the backbone LLM of FullStack-Dev. (3) FullStack-Bench, a comprehensive benchmark that systematically tests the frontend, backend and database functionalities of the generated website. Our FullStack-Dev outperforms the previous state-of-the-art method by 8.7\%, 38.2\%, and 15.9\% on the frontend, backend, and database test cases respectively. Additionally, FullStack-Learn raises the performance of a 30B model by 9.7\%, 9.5\%, and 2.8\% on the three sets of test cases through self-improvement, demonstrating the effectiveness of our approach.

Generating runnable front-end code from UI screenshots is a long-standing goal in automated software engineering. Existing MLLM-based methods predominantly focused on HTML/CSS, leaving multi-framework generation for React/Vue/Angular underexplored. Naively modifying prompts leads to substantial performance gaps across multi-framework and highly framework-specific error modes. To address this, we propose MulFCoder, a framework-conditioned multi-agent method that explicitly encodes framework constraints to bring multi-framework differences into a decidable rule space. MulFCoder orchestrates four agents: Grounder constructs an ElementTable, ContentTable, and macro-layout regions from detected UI elements; Planner builds a DOM-like hierarchical layout tree, produces a task schedule, and derives a framework-specific file Contract; Writer generates structured file writes or patches within a restricted edit window; Judger enforces lightweight, framework-conditioned constraints to accept or reject updates and trigger bounded repairs, preventing drift and deadlocks without expensive builds. Experiments demonstrate that MulFCoder substantially improves compilation success rate and reduces framework-specific errors, with particularly pronounced gains on Angular.

Despite recent progress, LLM agents still struggle with reasoning over long interaction histories. While current memory-augmented agents rely on a static ``retrieve-then-reason'' paradigm, this rigid pipeline design prevents them from dynamically adapting memory access to intermediate evidence discovered during inference. To bridge this gap, we propose MRAgent, a framework that combines an associative memory graph with an active reconstruction mechanism. We represent memory as a Cue–Tag–Content graph, where associative tags serve as semantic bridges connecting fine-grained cues to memory contents. Operating on this structure, our active reconstruction mechanism integrates LLM reasoning directly into memory access, allowing the agent to iteratively explore and prune retrieval paths based on accumulated evidence. This ensures that memory retrieval is dynamically adapted to the reasoning context while avoiding combinatorial explosion caused by unconstrained expansion. Experiments on the LoCoMo benchmark and LongMemEval benchmark demonstrate significant improvements over strong baselines (up to $23\\%$), while substantially reducing retrieval cost, highlighting the effectiveness of active and associative reconstruction for long-horizon memory reasoning.

While Large Language Model (LLM) agents have shown encouraging progress in financial reasoning, their effectiveness in realistic financial workflows is severely hindered by the lack of holistic benchmarks and the fragility of unstructured reasoning. To address this gap, we introduce ProFinR, the first Professional Finance Reasoning benchmark, covering four financial domain tasks and comprising 528 expert-designed problems. To solve these complex financial reasoning questions, we construct the Financial Tool Universe, a tool library containing 53 domain-specific tools organized into 13 categories. Building on the tool library, we introduce ProFinAgent, a structured agent framework based on Directed Acyclic Graph (DAG) and Case-Based Memory (CBM). Compared with strictly sequential workflows, ProFinAgent coordinates tool execution through a DAG planner for parallel efficiency and uses CBM to retrieve prior cases for more reliable decision-making. Experimental results demonstrate that ProFinAgent achieves a 49.81% performance gain over state-of-the-art baselines with a 47.1% reduction in inference latency. Our code is available at https://github.com/tsinghua-fib-lab/ProFinAgent.

Training autonomous web agents is fundamentally limited by the environments they learn from: real-world websites are unsafe to explore, hard to reset, and rarely provide verifiable feedback. We propose VeriEnv, a framework that treats language models as environment creators, automatically cloning real-world websites into fully executable, verifiable synthetic environments. By exposing controlled internal access via a Python SDK, VeriEnv enables agents to self-generate tasks with deterministic, programmatically verifiable rewards, eliminating reliance on heuristic or LLM-based judges. This design decouples agent learning from unsafe real-world interaction while enabling scalable self-evolution through environment expansion. Through experiments on web agent benchmarks, we show that agents trained with VeriEnv generalize to unseen websites, achieve site-specific mastery through self-evolving training, and benefit from scaling the number of training environments.