Agregador

arXiv:2610.22407v1 Announce Type: cross Abstract: In existing agent-based simulations of multimodal urban mobility, agents, representing humans, choose their travel modes either using discrete choice or machine learning models fitted on structured travel survey data, or with rules explicitly written by the domain experts. Such approaches fix a priori the agents' possible behaviour space. Hence, a factor absent from variables or rules cannot affect agents' decisions. In reality, choosing a transport mode is a multidimensional decision in which objective and subjective factors intertwine and change over time. Integrating such complexity into a city-scale multimodal mobility simulation remains a challenge. Generative agents (GAs) built on large language models promise to respond to this challenge because they carry decision heuristics that travel surveys do not record. However, to the best of our knowledge, aligning their decisions with real household travel surveys remains an open question. In this paper, we build an agent-based model of daily multimodal mobility for Toulouse (France). We compare the performance of fifteen decision-makers (simple baselines, models fitted on the survey, GAs with minimal and tuned prompts, and a typed classifier) on the real household travel survey. Our results show that GAs with tuned prompts come close to the fitted models on the modal split of a simulated day without transport disruptions, but not trip by trip. GAs also react when unforeseen events occur. The recently released Jev-1.13 TypeSafe classifier gives surprisingly great results, showing that generating output text is not necessary. The results presented in this paper would be of interest to transport policy makers and service providers looking to apply generative agent-based modelling for improving urban mobility services.

arXiv:2610.22382v1 Announce Type: cross Abstract: LLMs can generate Lean specifications from task descriptions, but those descriptions are often underspecified or ambiguous and may omit verifier-relevant edge cases, yielding imperfect formal specifications. We study whether an elicitation agent (EA) can improve verifier-checked Lean specifications by seeking clarification from a reference-grounded simulated developer before formalization. The agent decomposes the task, identifies unresolved decisions, asks bounded clarification questions, and uses the answers to write a refined natural-language specification that an LLM formalizer then translates into Lean. On 189 Verina Lean 4 specification-generation tasks, we compare single-shot generation with elicitation under question budgets of k=3 and k=10, allowing at most three or ten EA questions per task, across same-model and cross-model configurations. The same-model Claude Sonnet 4.6 pipeline has negative strict pass@1 estimates at both budgets, showing elicitation is not a uniform accuracy booster. In contrast, the Sonnet 4.6-to-GPT-5.6 and Sonnet 4.6-to-Gemini 3.8 Flash pipelines have positive strict estimates. The largest strict result reaches 65.6%, a +6.3-point change (+12 tasks), in the Sonnet 4.6-to-GPT-5.6 pipeline. Six of eight conditions have positive compile-conditional estimates, with a maximum cross-model difference of +4.9 points among compiling outputs. Treatment estimates also differ across the basic and advanced splits; the largest advanced-task change occurs in the Sonnet 4.6-to-GPT-5.6 configuration at k=10, from 42.0% to 50.6%. Failure analysis identifies postconditions as the largest source of degradation. These findings suggest future autoformalization systems should treat elicited natural-language specifications as intermediate artifacts and optimize jointly for semantic correctness among compiling outputs and reliable translation of clarified information into Lean.

arXiv:2610.22379v1 Announce Type: cross Abstract: Calibrating expensive engineering simulators requires tuning model parameters to match reference data, a process heavily constrained by computational costs. This study investigates whether Large Language Models (LLMs), conditioned on physical diagnostic descriptions of simulation errors, can propose informative parameter updates more efficiently than conventional derivative-free optimizers. We test this on a cylinder-wake Large-Eddy Simulation and a damped-oscillator inverse problem. Physics-labeled prompts are evaluated against four text ablations (removed, anonymized, padded, misdirected) and standard optimizers (Nelder-Mead, Bayesian optimization, CMA-ES, and random search). Results demonstrate a strong early semantic advantage: the first physics-informed LLM proposal improves the initial design in 23 of 25 trials. After five evaluations, the median calibration error is 21.9 times lower than physics-free controls. However, this advantage is transient - disappearing by fifteen evaluations - and inherent LLM decoder variability yields a 9.8-fold performance spread. To address this, we introduce a hybrid workflow that uses the LLM as a warm-start, passing accumulated evaluations to a conventional optimizer. This two-stage numerical continuation policy reaches target accuracy in just 9 simulator runs, outperforming the LLM alone (12 runs), Nelder-Mead (17 runs), and Bayesian optimization (>60 runs). Supported by sensitivity analysis validating the test cases, this hybrid physics-labeled LLM scheme proves highly effective as a calibration agent under strict evaluation budgets.

arXiv:2610.21016v1 Announce Type: cross Abstract: Large language model (LLM) agents for health can accumulate reusable textual skills from task feedback, such as calculation criteria, measurement time points, and formula units. Deciding which skills to retain is consequential: useful rules improve later answers, while faulty generalizations propagate errors. We propose CausalEvo, a method for causal skill self-evolution that estimates the relative effects of deliberately changed skill sets while holding the model weights, feedback cases, and rule count fixed. At each evolution step, a frozen LLM proposes candidate rules, which are pooled with the incumbent skills so that the current skill set remains a feasible choice. Paired skill replacements against shared background rules are scored on the same feedback cases; least squares converts the score differences into relative skill values, and the top-ranked rules are retained, favoring incumbents on ties. A decision-loss bound separates candidate support, approximation, and estimation errors. On six MedCalc-Bench-Verified tasks with 120 held-out cases, CausalEvo achieves 60.00% accuracy, compared with 55.83% for SkillOpt and 51.67% without skills. Across the recorded training configurations, which use different optimization budgets, it consumes 55.6% fewer training tokens than SkillOpt. Gains concentrate in particular calculations alongside regressions on others, and paired accuracy intervals include zero. CausalEvo turns task feedback into controlled decisions about skill retention and replacement. Project repository: https://github.com/txbrook/CausalEvo

arXiv:2610.23845v1 Announce Type: new Abstract: Reproducible and analytically robust research findings are essential to scientific progress, but evaluating these properties typically requires costly manual effort. We investigate whether LLM agents can help automate reproduction and robustness analysis, using a medical case study grounded in the Surveillance, Epidemiology, and End Results (SEER) database, a widely used U.S. cancer registry underlying tens of thousands of published studies. First, we introduce SEER-Repro, a benchmark of 20 published SEER-based papers together with the data and resources required to reproduce their findings. Second, we create an AI agent that successfully reproduces 77.4% of 2,190 atomic claims, while providing executable baselines for further analysis. Third, we develop an agentic analysis pipeline to test the robustness of the medical findings by identifying analytical choices underlying each finding, proposing defensible alternatives, and rerunning the analysis to determine whether those choices affect the conclusion. Across published papers in SEER-Repro, our pipeline identifies potentially consequential analytical choices in 9.5% of fully reproduced findings and flags them for expert review. These results show how LLM agents can support scientific review by reproducing findings, systematically testing their robustness, and directing expert attention toward potentially consequential analytical choices.

arXiv:2610.23826v1 Announce Type: new Abstract: Self-driving laboratories (SDLs) accelerate scientific discovery by automating experimental execution and optimization. AI agents can increasingly propose new experiments, but existing SDLs can only run the procedures they were built for. Adapting them to a new experiment still takes extensive manual engineering, which makes laboratory flexibility the bottleneck for scaling discovery. We introduce the self-building laboratory (SBL), which designs, configures, and operates its own laboratory setup given a high-level scientific goal. SBL starts from whatever already exists, from an empty room to a working SDL, and builds the missing capabilities: the lab layout, its digital twin, robot skills, and experimental workflows. We demonstrate the system working across different robots and setups, scaling from a single workstation to laboratories with several robots working together. We validate SBL's end-to-end capabilities through a real-world quantum-dot synthesis experiment. By shifting the role of human researchers from manually configuring and programming device automation to specifying scientific objectives, SBL makes scientific automation more accessible, flexible, and scalable. Project website: https://self-building-labs.github.io/

arXiv:2610.23801v1 Announce Type: new Abstract: Sparse autoencoders (SAEs) explain foundation model representations by decomposing activations into sparse combinations of learned dictionary atoms. However, the learned dictionaries can vary across training runs, limiting the reliability and reproducibility of SAE-based analyses. We propose Consistent Sparse Dictionary Learning (ConSDL), which replaces the standard linear encoder with per-sample sparse coding. This is motivated by our synthetic analysis showing that coefficients from a single linear encoder pass are suboptimal compared with those obtained by per-sample optimization, suggesting that sparse coding is a reliable mechanism for dictionary learning. Across 16 synthetic settings, ConSDL achieves near-perfect dictionary and coefficient recovery. These gains translate to state-of-the-art cross-seed consistency across four vision and language settings, where ConSDL improves dictionary and coefficient consistency by 17.6% and 35.9%, respectively, over the average of six SAE baselines.

arXiv:2610.23796v1 Announce Type: new Abstract: Transformer prefill reuses weights across tokens and is dominated by matrix multiplication and activation data movement; single-token decode has limited reuse and is dominated by weight and KV-cache reads. We present FUTRANSFORMER, a CUDA implementation of a Llama-style Transformer block for NVIDIA Hopper GPUs. It applies CODA's GEMM-plus-epilogue approach to inference, adding direct KV-cache placement, folding the fixed RMSNorm gain into weights, writing FP8 output directly with delayed scaling, and supporting prefill without padding. Decode uses INT4 projection weights and an INT4 KV-cache copy. Weights and keys follow Tensor Core fragment layouts for direct dequantization into MMA operands without full intermediate tiles. A block-diagonal Hadamard transform reduces cache quantization error from outlier channels. On one H100, prefill is up to 1.48x faster than compiled PyTorch FP8 and reaches 1.29 PFLOP/s. For gate/up projection with SwiGLU at the 8B shape, our BF16 path is 1.13x faster than CODA. Our FP8 path is 2.02-2.18x faster than the evaluated BF16 CODA path. Across 8B and 70B shapes at batches 1-32, INT4 decode is 1.3-2.4x faster than the faster of PyTorch's BF16 and INT4 blocks at each point: 2.1-3.3x faster than compiled, CUDA-graphed BF16 and 1.3-6.3x faster than the same block using PyTorch's INT4 projection kernel, which targets a single query row and loses to BF16 at batch 32. It is also 31-51% faster than our FP8 path. Complete Llama-3.1-8B decode takes 2.8 ms per step at batch one, versus 8.1 ms for the compiled transformers implementation. Decoding every WikiText-2 token through the INT4 weight path raises perplexity from 6.24 to 7.23 with round-to-nearest weights or 6.99 with GPTQ in the same format; the rotated INT4 cache adds 0.20. The kernels match a quantized reference computation within 0.001 perplexity.

arXiv:2610.23791v1 Announce Type: new Abstract: Computational drug discovery requires agents to compose scientific operations and adapt to changing evidence and constraints. Here we introduce DrugHarness, an evolvable agent harness coupling composable scientific execution with conditional skill induction. Its environment integrates over 80 scientific operations, biomedical resources and task-specific code, preserving associations among candidates, results and artifacts. From execution trajectories and outcome feedback, DrugHarness induces action hypotheses with observable applicability conditions, recommended actions and supporting evidence. An independent review agent evaluates candidate skills derived from these hypotheses through source verification and, when needed, targeted execution trials. Admitted skills guide subsequent tasks and are revised through feedback. Evaluation on MolBench and SMDD-Bench demonstrates capabilities in molecular screening, optimization and constrained design. Without evolution, DrugHarness exceeds reported MolClaw-CC results by 10.8 percentage points in affinity-comparison accuracy and 55% in mean top-three active hits on MolBench. Retrospective evaluation on skill-development tasks shows that evolution increases Qwen's affinity-comparison accuracy from 62.2% to 81.1% and optimization success from 79.5% to 87.2%. In a prospective HPK1 study, DrugHarness achieves a more favorable median docking score than Biomni among system-specific candidates. A synthesized compound achieves a cellular IC50 of 14.19 nM, a 2.64-fold potency improvement over the best input reference, with activity competitive with two selected compounds designed by experienced medicinal chemists. These results support conditional skill evolution for scientific agents and demonstrate a workflow connecting computational drug design with experimental validation.

arXiv:2610.23722v1 Announce Type: new Abstract: Understanding and simulating how subjective cognition and mental states emerge, evolve, and shape behavior is essential for advancing artificial intelligence toward deeper social cognition and general intelligence. Existing large language models (LLMs) and Theory of Mind (ToM) methods primarily infer mental states from linguistic information, lacking unified mechanisms for subjective world construction, cognitive-psychological interactions, dynamic evolution, and historical dependencies. This paper introduces the Mental World Model (MWM) and its implementation, the Multi-Branch Mental Evolution Architecture (MMEA), which explicitly constructs subjective worlds, models interactions between cognition and mental states, and simulates continuous mental evolution shaped by historical experiences. On DynToM and RecToM, MMEA achieves 96.48% and 94.00% accuracy, respectively, outperforming leading LLMs and specialized ToM methods. Additional experiments demonstrate improvements of 28.7, 39.3, and 16.6 percentage points in mental state and behavior prediction, long-sequence prediction, and implicit mental state inference, respectively. The results demonstrate that MMEA provides an interpretable framework for continuous mental modeling, enabling AI to move beyond surface-level semantic representations to infer individuals' underlying mental states from linguistic information and dynamically simulate their internal cognitive processes and evolution over time. MWM also provides a promising foundation for causal reasoning, personalized AI, sustained human-AI interaction, and future advances toward artificial general intelligence (AGI).

arXiv:2610.23670v1 Announce Type: new Abstract: Chain-of-Thought (CoT) monitoring offers actionable evidence for oversight AI systems, with the potential to enable early detection and intervention. However, existing approaches are focused on specific tasks, whereas the regulation of AI, particularly reasoning models, requires monitoring a broader range of tasks. We introduce Factor for CoT monitoring, a strategy that supports customizing monitoring requirements in the prompt, which enables a single monitor to cover different tasks by relying on different factors. Furthermore, we introduce two types of factors, fine-grained and coarse-grained (e.g., monitoring whether a specific email address has been leaked versus monitoring privacy leakage), to accommodate diverse user monitoring needs. Based on this, we develop a CoT monitor named OwlSight that relies solely on factor to monitor behaviors related to multiple dimensions, including capability, safety, and trustworthiness. Experiments across 15,971 reasoning trajectories show that OwlSight surpasses general-purpose LLMs by 1.32-30.57 and 11.71-27.81 points at fine and coarse granularity, respectively, and achieves state-of-the-art results on OpenAI's CoT monitorability benchmark. It also generalizes to unseen models' CoTs and remains robust to Factor rephrasings, CoT compression, and obfuscation, offering a promising paradigm for CoT monitoring that can contribute to the oversight of frontier AI.

arXiv:2610.23617v1 Announce Type: new Abstract: Identifying small-molecule structures from molecular formulas and tandem mass spectra remains a bottleneck in untargeted metabolomics and natural-product discovery. Even when database retrieval or de novo prediction misses the exact structure, the resulting candidates can provide starting points for exploring nearby structures through a few bond edits. We introduce GEMS (Guided Editing of Molecular Structures), which refines candidates through learned 2-switch bond edits that preserve the molecular formula and each atom's total valence. GEMS uses MIST, a pretrained spectrum-to-fingerprint model, to predict fingerprint probabilities once per spectrum. A neural policy compares each candidate's fingerprint with these predictions, links present features to the atoms and bonds that generate them, and scores all valid edits in one forward pass. The policy is trained on shortest edit paths between structures, so it can be pretrained on PubChem isomers without spectra and then fine-tuned with spectrum-predicted fingerprints. At inference, GEMS searches from multiple initial candidates matching the supplied molecular formula and ranks candidates by fingerprint agreement and multi-start consensus. With the target structure excluded from PubChem initialization, three edit rounds achieve Hit@1/Hit@10 of 29.64%/50.81% on NPLIB1 and 30.22%/41.33% on MassSpecGym, exceeding published baselines under comparable settings. On NPLIB1, GEMS achieves higher identification accuracy in about 3.3 minutes end to end, compared with 26.2 hours for our FRIGID run. GEMS also improves identification from generated candidates, including those refined by FRIGID. Experiments with exact target fingerprints identify fingerprint prediction error as a bottleneck in target recovery and final ranking. Deeper search recovers additional targets without retraining, with larger gains from weaker initial candidates.

arXiv:2610.23599v1 Announce Type: new Abstract: Large language model (LLM) agents tackle complex tasks through reasoning and interaction with external environments. Agent skills provide persistent, transferable capabilities that guide task planning, inform contextual decisions, support tool use, and enable executable procedures. Yet existing skill-evolution methods largely organize knowledge around skill artifacts, overlooking the functional heterogeneity within individual skills and its implications for knowledge maintenance and use. Inspired by cognitive science, we propose Cognitive Differentiation in Skill Evolution (CDSE), which evolves a unified skill through four functionally distinct yet coordinated components: Meta, Experience, Tool, and Script. CDSE extracts evidence-grounded semantic gradients from successful and failed trajectories, consolidates them by functional role, and applies component-specific updates and validation with cross-component coordination and adaptive invocation. An analysis of 500 top-ranked skill packages shows that 96.8% contain knowledge spanning at least two functional roles. Across six benchmarks, three agent models, and two execution harnesses, CDSE achieves the highest scores in all 36 settings, improving over the Base agent by 7.6 percentage points and the strongest competing method in each setting by 2.6 points on average. Ablation studies further support the importance of functional differentiation and adaptive invocation.

arXiv:2610.23598v1 Announce Type: new Abstract: Time-aware retrieval is crucial for long-horizon LLM agents to effectively leverage ever-growing long-term memory. Existing methods focus on lexical or semantic similarity, leading to target memories being crowded out by temporally mismatched yet semantically similar candidates. Even time-aware approaches inject temporal cues only during memory organization or answer generation, overlooking temporal constraints at the retrieval stage, where candidate selection occurs - the root cause of the displacement. Inspired by the temporal contiguity effect in human episodic memory, we propose TIDER (Temporal Interval-driven Dual-channel Evidence Retrieval), a time-aware agentic retrieval framework. Central to TIDER is a reinforcement learning (RL)-trained temporal interval locator that infers the temporal windows in which the evidence lies. RL optimizes window localization as a discrete, non-differentiable retrieval decision, using composite rewards to convert sparse answer supervision into fine-grained retrieval feedback. These temporal windows first define the temporal retrieval channel; the retrieved candidates are then merged with results from the semantic channel and reranked by proximity to the windows. This temporal grounding enables TIDER to retrieve temporally aligned evidence that would otherwise be drowned out by the volume of semantically similar yet temporally mismatched memories. With a lightweight 1.7B locator, TIDER achieves state-of-the-art accuracy on both long-horizon memory benchmarks - 61.52% on LifeBench (+8.82%) and 78.96% on LoCoMo (+3.66%) - which indicates that temporal cues are crucial for pinpointing target evidence amid a vast pool of semantically similar historical events in long-term daily-life memory.

arXiv:2610.23574v1 Announce Type: new Abstract: While Large Language Model (LLM) agents have demonstrated remarkable capabilities across complex tasks, they predominantly operate in isolated sessions, lacking the ability to continually learn from historical interactions or adapt to user-specific environments. Existing self-evolution paradigms often indiscriminately inject raw interaction trajectories into memory or skill modules, leading to obscured behavioral patterns and compromised update controllability. Furthermore, as foundational LLM reasoning improves, the bottleneck for agentic evolution shifts from accumulating generic knowledge to effectively managing situational experience. To address this, we propose Darwin-Claw, an environment-adaptive self-evolving framework that systematically distills episodic experiences to drive the comprehensive evolution of operational skills and durable memory. Theoretically grounded in an iterative agent-actor and evolver-critic co-adaptation, Darwin-Claw introduces a Hierarchical Experience Consolidation System that extracts fine-grained error attributions and generalizes them into dynamically updated prototype vectors. This structured situational experience serves as interaction patches for context-sensitive perception and explicit optimization signals for skill refinement. Practically, Darwin-Claw is a readily deployable, automated framework compatible with mainstream agent systems without requiring manual curation. Extensive experiments on WildClawBench demonstrate that Darwin-Claw yields substantial performance improvements, establishing a controllable and stable pathway for the continual, user-centric adaptation of autonomous agents in real-world ecosystems.

arXiv:2610.23554v1 Announce Type: new Abstract: Large language model agents increasingly rely on skills to extend their capabilities in complex tasks. However, scalable skill use requires Selective Skill Exposure (SSE): providing relevant procedural content when needed while avoiding unnecessary exposure that may disrupt reasoning. Existing paradigms fall short in complementary ways. Progressive disclosure exposes metadata for a growing skill library, introducing context pollution; retrieval limits catalog exposure, but static retrieval from the initial task query may be misaligned with capability needs that emerge during reasoning and may still inject unnecessary skills. We propose SkillBrowser, a Meta-Skill-managed framework that enables an agent to identify current capability needs, formulate targeted queries, and selectively load skills on demand. Across six SRA-Bench subsets and five LLMs, SkillBrowser improves aggregate accuracy over static retrieval and LLM-based selection, including multi-step tool use, and mitigates accuracy degradation on already-solvable instances. Benefits remain model- and task-dependent, with under-invocation of useful skills and additional latency as key limitations. These results demonstrate the viability of Meta-Skill-managed SSE for scalable skill repositories.

arXiv:2610.23525v1 Announce Type: new Abstract: Large language models (LLMs) are known to exhibit sycophancy by adjusting their responses toward users' expressed views. However, real users often justify their opinions through reasons, personal experiences, causal explanations, and broader narratives, raising the question of whether such narratives can further steer model judgments. We construct a controlled benchmark of 500 debatable questions across ten domains, where the user stance is held fixed while its justification is systematically varied without introducing new task-relevant evidence. Each question is instantiated under 11 matched conditions: one neutral baseline and five mirrored narrative levels for each of two opposing positions, yielding 5,500 evaluation instances. We evaluate seven LLMs and observe substantial differences in narrative-induced agreement. The effect is not monotonic with narrative elaboration: personal experience can reduce agreement, whereas causal framing and worldview-based synthesis increase it. Controlled GPT-5.1 ablations further show that causal framing produces the strongest positive effect, while claim-preserving bullet presentation also increases agreement over paragraph form. These findings suggest that sycophancy is shaped not only by user stance, but also by how supporting explanations are structured and presented.

arXiv:2610.23511v1 Announce Type: new Abstract: Locally deployed large language models (LLMs) can reduce reliance on remote cloud services for natural language smart building automation, but backend-facing processing requires correct outcome selection, alignment with local capabilities, and valid structured outputs. We investigate whether a complete outcome-specific prompting design improves compact-model outputs beyond generic route-then-generate prompting. The proposed LocalIntent approach routes requests to ordinary or persistent-rule branches: the former produces commands, clarifications, or rejections, while the latter generates trigger-action rules, followed by deterministic validation. We evaluate four locally executed Gemma and Qwen models spanning 2B-9B parameters on 101 non-persistent and 54 persistent-rule records, comparing Monolithic prompting, generic two-stage prompting, LocalIntent, and validation-guided repair. Across the evaluated datasets, LocalIntent achieves higher intent understanding, device identification, and action/parameter correctness than generic two-stage prompting, while clarification/mode correctness remains model-dependent. The largest gains occur for rules, where LocalIntent improves intent understanding from 26.4% to 80.6%, device identification from 59.3% to 84.7%, rule-mode correctness from 49.1% to 98.6%, and action/parameter correctness from 5.6% to 44.9%. For non-persistent requests, pooled gains are smaller: intent understanding increases from 39.6% to 53.2%, and action/parameter correctness from 30.7% to 42.8%.

arXiv:2610.23465v1 Announce Type: new Abstract: Graph Large Language Models (GraphLLMs) enable powerful semantic understanding and reasoning over Text-Attributed Graphs (TAGs), while their semantic reasoning may exhibit vulnerabilities to perturbations in node text and graph structure under adversarial attack evaluation. Existing attacks typically construct textual and structural perturbations in separate stages guided by distinct realization intents. However, this separation results in attack-intent fragmentation, causing semantic incoherence between the two perturbation types, thereby undermining attack stealth and making them easier for GraphLLMs to identify and disregard, ultimately weakening attack effectiveness. To avoid attack-intent fragmentation, we propose the Graph Attack Primitive (GAP), an atomic semantic representation that unifies textual and structural perturbations under a coherent attack intent. By carefully calibrating the semantic composition of a GAP, we can achieve strong attack effectiveness while maintaining sufficient attack stealth. Furthermore, we introduce S-GAP (Stealthy GAP), a multi-agent framework that iteratively refines the semantic content of the GAP. The framework unifies GAP planning and realization and continuously refines this process through surrogate evaluation and textual feedback. Extensive experiments across four datasets, three GraphLLMs, and two attack settings show that S-GAP improves ASR over the strongest baselines by 7.02% on average while producing more relationally plausible perturbations, exposing critical vulnerabilities in current GraphLLMs and motivating more robust defenses.

arXiv:2610.23449v1 Announce Type: new Abstract: AI agents have increasingly interacted with external environments by invoking tools, with Model Context Protocol (MCP) emerging as a standard interface for connecting agents to heterogeneous tools and services. As organizations deploy agents at scale, MCP gateways have become a critical middleware layer for aggregating, governing, and shaping the tool surface exposed to agents. Despite the growing importance of MCP gateways in agentic systems, no prior work has systematically studied how gateway design affects agent capability, efficiency, and security. To address this gap, we introduce GatewayBench, the first unified evaluation framework and benchmark for MCP gateways. We characterize gateway architectures into two major paradigms: flattened gateways and hierarchical gateways. GatewayBench comprises 524 tasks across six domains, together with a pool of 11,118 real-world tools from 251 authentic MCP servers. To emulate large-scale deployments, we systematically inject task-irrelevant tools sampled from this pool into each agent environment, enabling controlled evaluation as the accessible tool surface scales. Based on GatewayBench, we conduct extensive experiments across seven widely deployed MCP gateways and diverse agent configurations, covering both benign and adversarial tasks. Our results reveal substantial differences across gateway designs in task completion, tool-use efficiency, and resilience to attacks, and further demonstrate that these effects interact strongly with the underlying model and agent harness. These findings highlight the MCP gateway as a consequential yet underexplored component of the agentic AI stack. By enabling controlled evaluation of gateway architectures under realistic tool ecosystems, GatewayBench provides both practical guidance for MCP gateway design and a rigorous testbed for developing more capable, efficient, and secure agentic systems.