Overview

Web agents study how AI systems can carry out multi-step tasks on websites, from finding information to filling forms and making purchases. Common approaches combine visual or DOM-based page understanding with planning, action grounding, memory, and reusable skills; training often uses human demonstrations, synthetic environments, or reinforcement learning. Benchmarks such as WebArena, Mind2Web, and WebVoyager measure task success, while live-web tests expose challenges from changing interfaces and complex workflows. A central tension is improving reliability and efficiency without compromising security, privacy, or faithful execution of user intent.

Latest papers

189 papers25 papers in the last four weeks, up 47% on the four weeks before. 0.2% of all new papers.

Nov 13, 2026

WebArena-Pro: A Heterogeneous, Multimodal, Reproducible Benchmark for Web Agents

Web agents powered by large language and vision-language models are increasingly applied to realistic browser work that spans heterogeneous applications, multimodal content, and stateful workflows. However, existing reproducible web-agent benchmarks cover only a small number of web applications drawn from a few software categories, and restrict modality to text and vision. Live benchmarks broaden site coverage but sacrifice reproducibility, since pages and data drift between runs. Moreover, existing benchmarks do not meaningfully evaluate whether agents can understand and use audio and video content embedded within web tasks. To address these gaps, we introduce WebArena-Pro, a benchmark comprising 300 tasks across 20 self-hosted web applications in six domain categories, spanning distinct interface conventions, workflows, and data models. Across the evaluated agents, the best performance is achieved by both Codex with GPT 6.1 Sol and Claude Code with Claude Opus 5.5 at 62.3%, followed by OpenCode with Kimi 3.1 Pro at 60.7%. The strongest GenericAgent configurations, Gemini 3.1 Pro omni and Gemma4 31B, both achieve 30.3%. Among reproducible, human-curated web agent benchmarks, WebArena-Pro provides the broadest application coverage and the most comprehensive multimodal support to date. The benchmark treats audio and video as core observations alongside text and vision, with dedicated actions for extracting information from each. WebArena-Pro runs each task in isolation and supports reproducible, parallel evaluation. Tasks are authored through a dedicated annotator interface, filtered by LLM-assisted triage, and finally validated by humans before release. Project page: https://webarena-pro.github.io
Nov 11, 2026

GuardianAgentFlow: Multi-Agent Coordination for Web Privacy Protection

Web agents need to use private context without unnecessarily disclosing it. Effective protection must distinguish between requests that may be sent after sanitization and those that must be withheld, while preserving the information needed for authorized tasks. We present GuardianAgentFlow, a multi-agent framework that mediates outgoing Web requests through an explicit release contract. A dedicated restriction check grounds permission in the user's instruction; a shared protection scope aligns rewriting with disclosure checking; and candidate-grounded feedback guides revision before transmission. Experiments on Web requests with human-reviewed intent labels show improved request-contract success, while controlled comparisons and role ablations identify the contributions of restriction handling, policy evidence, and candidate checking. Evaluations of action proposals from PrivacyLens and AgentDAM demonstrate complementary protection when mediation is composed with privacy-aware agents. Experiments with smaller mediation models show that the coordination benefit persists with reduced model capacity. Dedicated-server measurements show lower mediation latency than full-policy reasoning, while real-site browser tests quantify integration overhead.
Nov 8, 2026

ScrapeBench: Evaluating Legal Compliance of AI Agents in Website Scraping

AI agents deployed on the open web are becoming more capable of performing actions that may breach contractual provisions and violate computer security laws. In this paper, we introduce ScrapeBench, the first agentic benchmark for empirically evaluating the legal compliance of AI agents with website scraping policies, namely terms of service (ToS) and robots..txt directives. Across widely used closed-weight and open-weight AI agents, we find that most agents exhibit low compliance rates (below 5%) when instructed to scrape websites in our sample of 1,626 popular websites that explicitly prohibit scraping. In addition, we observe that even where AI agents actively check the applicable ToS and robots..txt, they continue to exhibit low compliance rates. Conversely, in our sample of 269 websites that do not prohibit scraping, we observe that agents sometimes refuse to scrape despite it being lawful, such as Claude Opus 4.6 exhibiting an over-refusal rate of 98.5%. Our human baseline experiments (n = 180 participants) suggest that humans exhibit compliance rates comparable to AI agents but over-refuse at much higher rates.
Nov 4, 2026

MKD2World: A Typed Spatial MarkDown World Model For Web Agents

Graphical user interface (GUI) agents can improve decision-making with world-model foresight, provided that predicted states accurately reflect post-action interfaces. Existing state representations fail to jointly capture interface layout, UI semantics, and explicit locations. We introduce Typed Spatial Markdown (TSM), a Markdown-style textual state representation for desktop web interfaces. TSM serializes block-level and atomic-level UI elements according to their spatial relationships, jointly encoding their types, visual semantics, explicit locations, optional control states, and hierarchical membership. To train a screenshot-to-TSM renderer, we construct AtomBlock-WebUI, a dataset with precise annotations at both granularities. By injecting element-type labels into HTML and directly extracting bounding boxes after rendering, we avoid annotation errors caused by post-hoc inference from webpage structures. Building on TSM, we propose MKD2World, a desktop web world model trained in two stages for current-state description and action-conditioned next-state prediction. During inference, the web agent queries MKD2World as needed and performs a ``digest'' step, in which it interprets each predicted state and extracts task-relevant evidence for selecting the best action. Experiments show that MKD2World-9B achieves an EleF1 score of 26.50 on the out-of-domain test set, outperforming GPT-5.6-Sol(14.98) by 11.52 points. Across Online-Mind2Web tasks, MKD2World-9B improves the success rate of the Qwen3.5-35B web agent from 22.94% to 37.61%, achieving competitive performance among leading open-source GUI world models. Code is available at https://github.com/OmniAI-Lab-CUHKSZ/MKD2World.
Nov 3, 2026

ScamWorld: An Interactive Environment for Evaluating Web Agents Against Online Fraud

Web agents increasingly browse, shop, and manage accounts on behalf of their users. Prior work has shown that they can be misled by adversarial content, including prompt injections and deceptive interfaces. Beyond such attacks, agents operating on the open web also encounter online fraud designed to deceive humans, including fraudulent online stores, fake investment platforms, and advance-fee schemes. How web agents respond to such fraud during task execution remains underexplored, in part because scam websites are short-lived, making systematic evaluation difficult. To address this gap, we introduce ScamWorld, an interactive environment that reconstructs reported scam websites across diverse scenarios, fraud mechanisms, and languages. Each reconstructed website reproduces documented fraud mechanisms, making relevant signs of fraud observable before the agent's final action and enabling controlled, repeatable evaluation. Through an extensive study of web agents powered by six frontier LLMs, we find that baseline deception rates range from 11% to 35% across models. Agents proceed more readily with payments than with requests for credentials or wallet information. A closer analysis of agent trajectories shows that some agents acknowledge suspicious conditions yet proceed, at times treating the user's instructions as authorisation to accept additional terms imposed by the website. ScamWorld provides a reusable testbed for systematically evaluating and stress-testing web agents against real-world online fraud.
Nov 3, 2026

Interpretable Early Termination of Web Navigation Agents via Closed Sequential Pattern Mining

Web navigation agents built on open-weight language models fail often and fail late, with most of the tokens spent after the trajectory has become unrecoverable. Recent work has shown that early termination is a viable cost-reduction strategy, but existing approaches produce opaque termination decisions that do not support diagnosis or human oversight. In this work we apply closed sequential pattern mining (BIDE+) to K-step prefixes of agent execution traces to produce a small, human-reviewable failure-signature library that organizes into a four-category failure taxonomy with associated repair recommendations. We use a multi-variant evaluation protocol that explicitly removes contamination artifacts (zero-step crashes and trivially separable single-step successes) so that reported detection accuracy reflects behavioral discrimination rather than trace-length distributions. We evaluate our methodology on 1,544 MiniWoB++ traces collected from three open-weight language models, achieving the best F1 at K=3 on the Exclude-errors configuration across eight methods (0.717), recovering 88% of test-set failures within three steps, and supporting an early-termination policy that saves 26.3% of tokens at 92% precision.
Nov 2, 2026

WebHarbor: Docking Real Websites for Evolving GUI Agent Environments

GUI web agents require realistic and interactive environments that remain stable during both evaluation and training. Live website evaluation suffers from changing content, access restrictions, and limited coverage of authenticated workflows, while existing synthetic environments for training often lack functional depth and visual fidelity. We propose WebHarbor, a suite of 153 executable and dockerized website environments that faithfully reconstruct the visual interfaces, content, and functional workflows of real-world web services under deterministic control. We build WebHarbor through task-driven human-agent collaboration, where environments evolve as new tasks expand their functional scope, coding agents implement the required changes, and human contributors audit visual fidelity, functional correctness, and task validity. The resulting suite comprises 3,518 tasks spanning multi-step and post-authentication workflows, where existing works often fail to cover. Beyond reliable evaluation, WebHarbor provides high-fidelity synthetic web environments for cost-efficient reinforcement learning (RL). We demonstrate that a Qwen 4B-level GUI agent trained with RL in WebHarbor generalizes to unseen tasks and performs competitively with leading open-source GUI web agents across four benchmarks. The code and data are publicly available at https://github.com/aiming-lab/WebHarbor.
Nov 2, 2026

From Local Skills to Long-Horizon Web Tasks: Progressive Skill Exploration for LLM Web Agents

Large language models (LLMs) are driving autonomous agents toward real-world environments, with the Web emerging as one of the most representative interaction scenarios. However, training Web agents with naive exploration remains challenging: successful trajectories are rare, and the sparse terminal rewards make it difficult for agents to obtain useful feedback from failed trajectories. Inspired by skill acquisition theory, we propose Skill-Composed Progressive Exploration (SCOPE), a short-to-long learning framework for long-horizon Web agents that decomposes complex task exploration into a gradual expansion process from local skills to complete task-solving capability. Specifically, SCOPE iteratively explores around local skills: Skill Discovery extracts reusable local skills and failure experiences from historical rollouts; Skill Orchestration composes discovered skills into high-level skill chains; and Skill Expansion leverages failure experiences aligned with the current skill chain to guide the agent toward exploring new subskills. Through this process, SCOPE progressively constructs task-completing skill chains across rollouts and internalizes them into the policy parameters through agentic on-policy distillation. Extensive experiments on WebArena, WebVoyager, and WebShop show that SCOPE consistently improves exploration dynamics and achieves 3.9% to 18.6% relative gains over the strongest baselines in each setting.
Oct 31, 2026

When 200 OK Is Not an Answer: Silent Fetch Failures in Agentic Web Content Acquisition

Web agents and answer engines fetch pages at question time, and liveness checks of cited sources treat an HTTP 200 response as success. We measure how often that assumption fails and what it does to measurements built on fetched pages. We fetched 10,307 URLs in three days with a polite static client and a rendered browser client: pages from 1,507 sites in five strata, random non-existent paths on every site, and the 4,795 URLs cited by expert-verified claims in ExpertQA. We define a taxonomy of what a 200 response contains, label the static 200 responses automatically, and check the labels against 300 items reviewed by an LLM-based assistant; a human check of 60 of them retained every reviewed usability label after discussion. By the automatic labels, 20.9% of the pages that a static client receives with status 200 do not contain the page's main content: app shells, bot challenges, block pages, error and placeholder pages, and partial content. On the held-out test split, the reviewed sample lowers this estimate from 25.5% to 20.0%. 13.1% of the sites whose random-path probe was answered returned 200. The error propagates: for the cited sources, treating 200 as content overstates the share of claim-source pairs that can be checked against live content by 12.2 points and inflates the apparent silent citation-drift rate from 5.8% to 21.9%. Cheap heuristics catch most of it. Text length, probe comparison and redirect-to-root, combined and tuned on held-out sites, recover 86% of unusable reviewed responses at a false-alarm rate near 5%. Content-reading detectors perform comparably: the best, the text classifier, gains five to seven points of recall at more than twice the false-alarm rate. We argue that agents and their evaluations should report fetch outcomes explicitly instead of inferring them from the status line and will release the labelled data and code.
Oct 29, 2026

DynaRev: Benchmarking Web Agents under Dynamic Goal Revision

Web agents are typically evaluated with goals fully specified before execution and fixed thereafter, whereas real users often revise their requirements during interaction. To study such interactions, we introduce DynaRev, a benchmark of 181 tasks to evaluate web agents under dynamic goal revision. DynaRev uses semantic milestones to deliver revisions at task-defined stages, covers additive revisions, unconditional overrides, and runtime-conditioned overrides, and evaluates correctness, responsiveness, and efficiency through trajectory-grounded evaluation. Five-model evaluations show that agents often satisfy only part of the task requirements, with performance varying across task sources, revision types, and models. Further analyses reveal varied milestone timing, mixed effects of dynamic delivery, and contrasting revision-handling outcomes, with gains from instruction rewriting while preserving browser state and losses from browser resets. DynaRev provides a testbed for studying how web agents adapt to evolving user goals. The code and data are available at https://github.com/zsq259/DynaRev.
Oct 29, 2026

AgentHeal: Evaluating Goal-Preserving Resource Recovery of Web Agents under Access Barriers

Large language model (LLM) agents have demonstrated strong capabilities in automating complex information-seeking and transactional tasks across the open web. However, their real-world reliability remains limited by access barriers, such as dead links, authentication walls, rate limits, and anti-scraping defenses. These barriers block the primary execution path, causing agents to abort prematurely, loop indefinitely, or quietly compromise the user's original constraints. Existing evaluations either implicitly assume that target resources are freely accessible or focus on transient perturbations that simple retries can resolve, failing to capture scenarios where agents must actively reroute around impassable bottlenecks. To bridge this gap, this paper introduces AgentHeal, a comprehensive benchmark evaluating goal-preserving resource recovery, defined as the ability of web agents to detect non-transient access barriers, pivot to legitimate alternative sources, and fulfill the user's exact original objective. Spanning nine barrier families across a counterfactual sandbox and real-website replay, our experiments on leading LLM backbones show that average task success drops sharply from 89.4% to 43.7% under access barriers. Further analysis reveals that agents frequently fall for deceptive look-alikes, fail to escalate to users when necessary, violate access policies, or waste computation on futile retries, providing actionable diagnostics for developing resilient web agents.
Oct 27, 2026

WebAnything: From the Wild Web to Scalable Agent Environments

Web agents require diverse, executable, and resettable website environments for reproducible training and evaluation. Human-assisted mirrors demand per-site engineering, while generative construction can diverge from specified real websites. Capture-and-replay preserves observations, with interaction coverage constrained by captured content and replay capabilities. We present WebAnything, a fully automated framework that reconstructs source-grounded, reusable Web environments from real URLs without per-site manual implementation or repair after initial setup and authorization. An Explorer-Builder collects browser-observable structures, resources, and interaction outcomes; an independent verifier checks frozen candidates and guides local repair or targeted source recapture. We construct 244 website artifacts across 14 application domains and 53 categories, supporting 12,984 executed and accepted tasks. Compared with human-assisted WebHarbor, WebAnything achieves higher aggregate scores on three fidelity metrics, with 12 wins, two ties, and no losses in a blind audit. Fine-tuning Qwen3.5-9B on 5,856 interaction trajectories improves REAL success by 9.9 percentage points, Odysseys-Easy Rubric Avg and Task Avg by 6.25 and 6.32 percentage points, and WebArena-Verified Overall success by 13.33 percentage points under the fixed harness. Code: https://github.com/china-qijizhifeng/Web-Anything
Oct 26, 2026

Back to Basics: A Skill-Centric Approach to Generalist Web Agents

Browser agents do not just need more practice. They need to practice the right things. We introduce MiniWeb, a realistic miniature of the web comprising 65 interconnected websites across 15 domains, where agents navigate, search, and transact as people do. Its 369 human-annotated tasks cover everyday workflows and label 37 reusable interaction primitives grouped into 7 skill families, allowing us to diagnose which skills fail rather than only whether a task succeeds. Benchmarking ten open-weight agents, we find that drag & gesture is consistently the weakest family, with no agent solving more than 24% of its instances. To address this gap, we introduce MiniWebGen, a framework that autonomously generates verified, skill-centric training data from MiniWeb. Fine-tuning Qwen3.5-4B on just 3.1K verified trajectories improves single-primitive success by 24.9pp across 13 unseen websites, with gains across every skill family. However, these trajectories teach short-horizon interactions, which do not directly translate to full-task success. We build on this result with MiniWebAgent, which pairs the trained executor with a planner that decomposes full tasks into the skill-level steps it learned to perform. This approach improves task success by 19.4pp points on unseen MiniWeb sites and transfers to external benchmarks, with gains of 9.3pp on WebArena-Lite, 16.1pp on Online-Mind2Web, and 20.0pp on WebVoyager, without training on their trajectories. We provide a simple recipe for compact browser agents: practice reusable skills directly, then compose them to solve unfamiliar tasks.
Oct 26, 2026

Proof-of-Agentness: Towards Verifying Agents on the Web

CAPTCHAs and bot detection have helped protect Web services from automated abuse for decades by distinguishing humans from bots. However, the emergence of LLM-powered agents introduces a new setting: Web services increasingly accommodate agents as legitimate users. Without a way to distinguish agents from other automated clients such as simple bots and scripts, these services risk large-scale automated junk submissions that waste service resources and consume users' attention or tokens. In this paper, we investigate a novel question for today's Web: "How can a Web service obtain evidence that a client is an agent?" We introduce the concept of "Proof-of-Agentness" (PoA), a mechanism for distinguishing agents from other automated clients, motivated by the parallel between human attention and an agent's inference resources. We outline two approaches: active verification through challenge-response and passive analysis of service-visible signals. Eight language models achieve an overall accuracy of 96.5% on our challenges, with an average overhead of 1,647 tokens. For passive analysis, we conduct an IRB-approved user study comparing the Web interactions of twelve human participants and representative Bot and Agent implementations, achieving an average classification accuracy of 95.8% across four classifiers, motivating further study.
Oct 25, 2026

LLM WikiLeaks: Uncovering URL-Based Capability Extension and Security Circumvention in Agentic Message Boards

Autonomous agents have been observed acting beyond their assigned tasks, accessing unauthorized systems, and establishing communication channels to exchange information, yet the Web and network-level techniques they use to extend their capabilities and circumvent security restrictions remain poorly understood. We study these techniques through the ProWiki incident, in which agents restricted to outbound HTTP GET requests repurposed a public German-language wiki as a message board. We analyze 14,591 agent edits containing 79,541 URL postings and present corroborating evidence that the published URLs reflect the agents' operational activity. We find that 26,717 URL occurrences (33.6%) embed target URLs within one to five layers of third-party Web services. By recursively reconstructing these nested URLs, we identify 67 distinct intermediate services and classify their functionality. Most intermediate layers (73.8%) extend agents' capabilities by delegating content extraction and transformation to external Web services: Markdown and OCR services convert web content into machine-readable text, JSON filtering services extract selected fields, and archival services expose historical versions of pages. The remaining 26.2% facilitate circumvention of access restrictions and evasion of detection, including CORS and fetch proxies that retrieve content on agents' behalf, and link shorteners and redirectors that hide destination URLs from mechanisms inspecting literal URLs or outer hostnames. Agents also reach the same targets through multiple URL paths, consistent with attempts to evade rate limits or obscure request origins. Our results expose a gap between the capabilities granted to agents and those they can obtain by composing publicly available Web services: restricting an agent's local toolset or direct Internet access may be insufficient while third-party services remain reachable.
Oct 24, 2026

Agent-to-Agent Synthetic Social Engineering on the Web

An AI agent can prepare an accurate report and still request an action its user has not authorized. Cooperation across Web services therefore requires separate judgments about identity, evidentiary support, and permission. We define synthetic social engineering (SSE) as deliberately deceptive or misleading influence intended to exploit AI recipients' responses to credibility, authority, or obligation cues and redirect behavior away from authorized objectives. We focus on separately tasked sender agents that produce or adapt misleading communication. Drawing on cybersecurity, social engineering, and persuasion research, we develop an operational framework connecting actors, deceptive representations, candidate influence routes, evidence or permission boundaries, responses, and consequences. The framework distinguishes attempts from successful redirection, without assuming human psychological processes. We implement it in the Agent Service Exchange and Evidence Commons (ASE-EC), a reproducible testbed with independently specified records and authorization scopes. Across 1,260 research episodes using Qwen3-4B in two fictional briefing tasks, matched comparisons test authority and commitment appeals, while defense comparisons assess protection alongside authorized task completion. Invoking a limited prior agreement increased unauthorized requests from 43.3% to 96.7%, although recipients retained supported decisions and permissions remained unchanged. These controlled findings identify action selection as a distinct security concern, even when evidence handling succeeds, and they provide a reusable basis for evaluating safeguards that preserve useful cooperation across Web services.
Oct 23, 2026

WebUniverse: Scaling Verified Training Data for Agentic Web Development

Training agents for front-end web development requires demonstrations of code editing, execution, browser inspection and revision, yet public trajectories and reusable task environments remain limited. We introduce WebUniverse, a pipeline that constructs verifiable tasks with verified development trajectories from open-source repositories and real web pages. The tasks cover text-grounded application development in standalone and iterative settings and vision-grounded reproduction across pages and viewports. We collect agent trajectories for these tasks and verify the resulting applications against task-specific functional or visual requirements. We release WebUniverse-Env, containing 15K executable task environments, and WebUniverse-Mix, containing the corresponding 15K development trajectories with verification records and scores. Fine-tuning Qwen3.5-35B-A3B on WebUniverse-Mix raises the Vision2Web average score from 51.7 to 68.3, WebGen-Bench accuracy from 34.5% to 63.9%, and Web-Bench pass@2 from 12.2% to 23.1%. The resulting model also outperforms a public-data baseline trained with the same configuration and sample budget across all reported metrics. Ablations show complementary benefits from text and visual trajectories and continued gains as training data scales from 5K to 15K trajectories.
Oct 22, 2026

Learning What to Ask for Tomorrow: Active Preference Acquisition for Long-Term Personalized Web Agents

Web agents increasingly serve the same user across repeated online tasks such as booking travel, shopping, and scheduling, so their long-term usefulness depends on how well they know the user's preferences. Passive memory and personalization methods cannot obtain preferences the user never reveals, and clarification methods that value a question by its benefit to the current task overlook the benefit of reusing the answer in later tasks. We study Active Preference Acquisition (APA): which missing preferences are worth the cost of asking? We define the Prospective Personalization Value (PPV) of a missing preference as its expected cumulative utility gain over future tasks minus the effort and disclosure-risk cost of asking, and select questions under a per-interaction budget. We characterize when cumulative reuse justifies acquisition despite insufficient single-task benefit, and evaluate this regime with an initial-profile diagnostic. APA computes the counterfactual gain of each candidate preference on public task prototypes with the agent's own task solver and aggregates it by the expected reuse of each task family. We evaluate it on ActiveWebPref, a controlled longitudinal benchmark with hidden but queryable user profiles and exact task utilities. At a horizon of 20 tasks, 34.4% of budget-feasible valuable missing preferences are amortization-dependent; APA acquires 80.3% of them, whereas current-task value of information acquires 5.1%. Under per-user asking-cost caps matched to APA's realized expenditure, APA achieves higher net utility than four heuristic selection baselines. In our cost setting, asking about every unknown preference yields lower net utility than not asking, while APA recovers 84.2% of its utility gain with 27.6% of its questions. The value of acquiring a user preference should account for its reuse across future tasks, not only its benefit to the task at hand.
Oct 21, 2026

CORD: Conflict-Aware Optimization for Joint RL and Distillation in Agents

Joint reinforcement learning and self-distillation combines task rewards with dense teacher guidance for agent training, but their gradients can partly cancel. A mixing coefficient controls relative scale without selectively removing opposing components. We present CORD, a general framework for conflict-aware optimization of two objectives. CORD separates the choice of protected objective from the condition that triggers projection, and combines the gradients once per optimizer mini-batch. A nearest-point formulation yields a closed-form update and characterizes its first-order effect on the protected objective before optimizer transformations. Gradient diagnostics connect conflict in agent training to lost local optimization progress and motivate selective correction. As a gradient-combination mechanism, CORD can be added to different joint training recipes while retaining their objectives. Across three such recipes, it improves ALFWorld success by 1.5-15.9 percentage points. On WebShop, it improves two of the three pairs, with SDAR rising from 60.7% to 64.6% success. Our analyses explain why protection direction and projection gate must be chosen together, and why the distillation objective and its mixing weight remain consequential after projection.
Oct 19, 2026

LoWAgent: A Low-Resource Multilingual Benchmark for Web Agents

Web agents are typically evaluated on English-language interfaces, leaving their ability to operate across the world's multilingual websites largely unexplored. This gap is especially significant for low-resource languages, where standardized resources for evaluating web agents remain limited. To address this gap, we introduce Low-Resource Web Agent Benchmark (LoWAgent), a benchmark for evaluating web agents across different low-resource languages (LRLs) in comparison to the default, English. LoWAgent evaluates web agents across English and four LRLs: Hindi, Bengali, Urdu, and Amharic. The benchmark comprises of six fully localized websites, instantiated across five languages to yield 3,000 task instances. We evaluate three closed-source and five open-source models across all five languages on all six websites. Our findings show that both closed-source and open-source models experience performance degradation for LRLs compared to English, with substantially greater performance drops observed among the open-source model agents. We further evaluate the robustness of the models on realistic interface disruptions, such as cookie notices and advertisement pop-ups, and find that these further degrade agentic performance. Finally, we show that post-training on a small subset of LoWAgent traces can significantly improve low-resource language agentic task performance, demonstrating the usefulness of our benchmark.
Oct 19, 2026

RefCraft: Benchmarking Visual-Reference-Conditioned Agentic Generation through Web Development

Beyond basic tool use and coding skills, effectively incorporating visual references into agentic generation is equally essential for applying AI agents to real-world productivity tasks. We study this crucial capability through webpage development, a practical and readily evaluable setting in which existing benchmarks remain largely idealized: they typically focus on reconstructing visual targets or generating webpages from textual instructions alone, overlooking the more common workflow of developing from visual examples. Specifically, we introduce RefCraft, a benchmark comprising 300 realistic multimodal webpage-development tasks that combine natural-language instructions with screenshots from one to five reference webpages. Agents must selectively interpret and utilize these references while rendering, visually inspecting, and iteratively refining their outputs. We evaluate the resulting open-ended artifacts using independently verifiable keypoints and complementary programmatic and VLM judges. Across nine frontier (M)LLMs, the strongest system achieves an overall score of only 52.12 out of 100, revealing that visual-reference-conditioned agentic generation remains a substantial challenge for current models in practical webpage-development settings.
Oct 15, 2026

Don't Retrain, Remix: Adaptation of Web Agents using Mixture Hypernetworks

The web is not static or uniform. Websites that serve the same purpose can look and navigate very differently. Even within a single site, successive redesigns can change how familiar features are presented and accessed. Humans adapt to this by recognizing familiar interaction patterns and reusing what they already know. We give web agents a similar ability with WebMix, a mixture hypernetwork that adapts them to unseen interfaces by composing previously trained low-rank adapters. WebMix begins with a bank of adapters trained on source interfaces. At deployment, a scout agent explores the target interface and summarizes its observations into a compact manual. A lightweight router uses this manual to predict mixing coefficients over the adapter bank, which combine the existing adapters into a new one for inference. Adapting to a new interface takes only a scouting run and a forward pass, without gradient updates on the target. With Qwen and Llama models, we evaluate domain transfer on TimeWarp and WebArena-Lite-v2, and version transfer on TimeWarp. In all three settings, WebMix matches or outperforms in-context learning and LoRA baselines, often matching the average success of specialists trained directly on those interfaces. On TimeWarp domain transfer, it gains about 8 percentage points on average over the best baseline with both models. These gains do not come at the cost of seen interfaces, where WebMix keeps its lead, and extend to version transfer. Adapting to a new interface can therefore rely on exploration and composition instead of retraining.
Oct 14, 2026

STP-GUI: Spatially-Aware Token Pruning for Efficient Web-GUI Agents

With the rapid development of vision-language models (VLMs), Web-GUI agents have achieved promising performance in interacting with real-world interfaces through screenshot-based grounding. However, high-resolution screenshots and multi-turn histories introduce excessive visual tokens, causing substantial inference overhead. Existing token pruning methods, mainly designed for natural images, often select tokens independently based on importance scores, which can fragment the local visual structures required for recognizing small GUI elements such as buttons, icons, and hyperlinks. In this work, we propose STP-GUI (Spatially-aware Token Pruning for GUI agents), a query-guided token pruning framework that preserves fine-grained instruction-conditioned token scoring while performing spatially coherent window-level selection under aggressive token reduction. A lightweight selector is optimized through policy learning while the VLM backbone remains frozen. Experiments on Online-Mind2Web show that \papername achieves a favorable efficiency-performance tradeoff under low token budgets. At 21% token retention, it preserves 77.7% of full-token performance while reducing inference latency by 31% and FLOPs by 57%. Results on Android in the Wild further suggest that the method transfers to mobile GUI environments.
Oct 14, 2026

Why Generate a Click? Plug-in Decision Heads for Web Agents

Web agents built on vision language models decide every action by generating text such as click(12), although most of that text encodes a choice among a few action types and a few dozen marked elements. We read these choices directly from the hidden states that the generator computes while it reads the goal and the screenshot. Plug-in decision heads on a frozen generator predict the action type, pool the visual tokens inside each numbered box to pick the target, and point at a listed text option or at a WRITE slot that leaves the writing of new text to the generator. A confidence cascade executes the heads' action when they are sure and otherwise lets the generator decode from the KV cache of the same forward pass. With Qwen3-VL 4B and 8B generators trained on 213k web steps, a strict threshold keeps the cascade between 0.3 points below and 1.4 points above the generator on seven offline test sets while the heads decide 12 to 64% of the steps. A looser threshold raises step success on unseen Mind2Web websites by 3.1 and 1.8 points, costs up to 2 points on GUIAct and MiniWoB++, and lifts WebArena task success from 12.5 to 13.8% (4B) and from 13.6 to 14.6% (8B) with less than half of the generated tokens. Training the heads jointly with the generator lowers the generator's own step success by up to 4.9 points; heads on a frozen generator avoid this cost.
Oct 14, 2026

Teaching Agents in their Own Style: Blackbox Teacher Distillation for Web Agents

Frontier web agents using closed models have high inference and computational costs. Smaller open models can be an alternative, but to be performant, they require training supervision from a human or teacher model. Token-level teacher distillation is not an option for closed teachers as it needs access to the model's logits. The standard approach is to imitate teacher demonstrations offline, but it trains the student on states it has never visited, and in a style it does not write in. We introduce WebKD, where the teacher is the editor of the student's failures. Given a failed episode, the teacher finds the first point of mistake, and rewrites the student's reasoning traces, harness states, and actions from that point onward. The editing reveals that the teacher's corrections can be more surprising to the student than the student's own text. WebKD instead aims to edit minimally. In the data space, the teacher writes each correction in the student's style. In the objective space, the reasoning is weighted by its contribution to the action. A role-asymmetric KL anchor allows the corrected action to move freely while keeping the reasoning near the student's prior. On the TimeWarp benchmark, WebKD students gain about 5.5-10.7 pp over behavior cloning, outcome-reward RL, and teacher editing baselines. The performance transfers to unseen versions and to the live web in the OnlineMind2Web benchmark. WebKD opens a new way of distilling black-box models to improve the capabilities of open-weight web agents.
Oct 12, 2026

Get In, Decide, Prove: A Merchant-Side Readiness Benchmark for Agentic Commerce

AI agents now browse and buy on people's behalf, and benchmarks measure how well the agents perform. Far less is known about the other side of the transaction: whether a merchant's website lets an agent in, publishes terms precise enough to decide a purchase, and leaves evidence of what was agreed. This paper presents a merchant-side benchmark built on those three questions and reports a pilot of it. The central measurement, decision sufficiency, rates each criterion of a concrete buyer mandate (total cost, cancellation or return terms, service date, variant identity) as determinate, indeterminate, or unknown from the merchant's own published evidence. Readiness scores carry their evidence coverage and missingness bounds; access is reported separately for a declared bot, a declared headless browser, and a user-present browser; and each observed term is mapped to the RAILS admissibility class of the evidence that could support it in a dispute. Four large merchants received the full assessment: three refused the declared profiles at their offer pages, and no mandate could be confirmed as met, because each merchant left a total or a refund term undetermined. An access census of 555 sites of merchants that two payment processors publicly name as clients found the home page unreadable to both declared profiles at 105 sites (18.9%). A declared assessment of 330 merchants, scored by deterministic rules without a rater, found that evidence coverage rather than the score decides who can be ranked: 115 merchants (35%) reached the ranking threshold, rankable profiles scored a median of 89 to 90, and material terms was the one pillar well below the others, because the all-in total was decidable at 18 merchants. The pilot shows that the measurements can be taken; it does not validate them or rank merchants, and the weights and the projected clearing band await validation against outcomes.
Oct 12, 2026

HyWe: Unifying Search and Operation for Real-World Hybrid Web Tasks

Real-world hybrid web tasks need both information gathering and page interaction, yet search-based and operation-based agents are developed and evaluated separately. To bridge this gap, we introduce HyWe, a framework that unifies both capabilities within one model and trajectory. Its two modes use dedicated prompts and tool sets to sustain sub-goal execution; explicit switching selects the needed capability, and shared history lets each mode build on the other's findings. To provide demonstrations of these behaviors, we develop HyWeCollector, a fully automatic pipeline that uses task prototypes, four switching patterns, and multi-agent synthesis to produce verified hybrid trajectories. We train HyWe-35B on these demonstrations to unify efficient information gathering and browser interaction in a single 35B web agent. To evaluate hybrid task completion, we introduce HyWeBench: 100 human-written tasks on websites distinct from training, with human-written sub-goal checklists and an outcome-based evaluation pipeline. Hybrid training enables HyWe-35B to achieve 47.0% success on this benchmark, improving on the untrained base model by 19.0 percentage points and outperforming existing search-based and operation-based agents. Even before training, the same base model solves 28.0% of tasks with HyWe, versus 11.0% with all tools under one prompt. Beyond hybrid tasks, HyWe-35B achieves 59.0% and 88.3% success on Online-Mind2Web and WebVoyager, outperforming existing open-source baselines and cutting action cost per successful task by 48.3% and 51.9% compared with Fara-7B.
Oct 9, 2026

A Baseline Assessment for the Agentic Web Technology Stack

The fast pace in which protocols, frameworks, and technologies are being proposed for LLM-based agents results in a complex scenario for anyone aiming to create state-of-the-art AI solutions. Without an assessment on how different protocols, frameworks, and libraries work together to compose an Agentic Web solution, developers and organizations face different challenges (e.g., skilling, governance, infrastructure) even before becoming part of the Agentic Web ecosystem. In this context, we propose a system for performing Agentic Web evaluations called MASTIF (Multi-Agent System TestIng Framework) and present a baseline assessment of Agentic Web technologies, including 3 LLMs, 3 protocols, and 4 agent implementation frameworks, against 300 Mind2Web benchmark tasks. Results suggest that selecting a task-specific technology stack may result in gains of task completion rate, efficiency score, and social value stability scores; and reductions in verbosity and time to complete tasks. With the presented results, we expect to present a baseline in terms of metrics and results for this complex and fast evolving scenario for those aiming at designing, assessing, and developing multi-agent systems.
Oct 9, 2026

Web$hell: A Unified Interface for Agentic Web Access

Large language model (LLM) agents are becoming direct users of the open web. As they take on increasingly complex and information-intensive tasks, effective web access is becoming a core capability for agentic problem solving. Existing approaches have advanced agentic web access through different access models, including search-augmented agents, browser-based agents, and, more recently, filesystem-based agents. Nevertheless, as users expect agents to complete long-horizon, end-to-end tasks, they increasingly need to conduct broad information gathering, perform browser actions, and revisit acquired evidence within a single trajectory. Yet our preliminary analyses reveal that existing access models struggle to meet these demands. They primarily support either information gathering or browser interaction, while combining their capabilities can involve substantial costs or performance trade-offs. To this end, we present Web$hell, an interface that unifies web access under a single access model. Using a common schema defined by WebContract, WebCompiler compiles live webpages into addressable information and action objects. These objects are maintained in WebCache and handled through a shared set of operations, providing agents with consistent, operable, and persistent web access. Through extensive experiments across web search, web navigation, and deep research benchmarks, we show that Web$hell achieves strong task performance at lower cost. Our results further indicate the importance of interface design for effective web access, with Web$hell opening new possibilities for more capable and efficient LLM agents on the web.
Oct 9, 2026

When Charts Deceive Web Agents: A Paired Benchmark for Misleading Visualization-Grounded Tasks

Web agents increasingly browse, interpret, and act on web content on behalf of users. In many realistic workflows, however, the visual evidence they consume is not neutral: charts and dashboards can be misleading, shaping judgments without issuing any explicit malicious instruction. Unlike chart question answering, Web Agents treat charts as intermediate evidence whose misinterpretation can propagate into concrete downstream actions --- a reliability risk that remains underexplored. We introduce MisVisAgentBench, the first paired executable benchmark for evaluating how misleading visualizations affect Web Agent execution. It comprises 280 task instances in matched clean--misleading pairs that differ only in chart evidence. We further provide a diagnostic framework that isolates misleading-induced failures from baseline task difficulty and captures execution-level effects such as extra browsing steps and decision hesitation. Across thirteen vision-language Web Agents, replacing clean charts with misleading counterparts reduces the average task success rate by more than 34 percentage points. Moreover, these failures predominantly follow the targeted misleading action branch, suggesting that deceptive charts systematically redirect rather than randomly disrupt agent decisions. These results reveal that visualization robustness is a critical yet understudied dimension of Web Agent safety and reliability.