# agentic browsing

Agentic browsing is a web-development concept involving AI agents interacting with websites and technical audits that evaluate machine interaction readiness.

This is one page of public article previews, not the complete archive. Follow Next page to continue. Summaries are not the original full articles.

## Securing Your Newest Customer Channel with Auth for Personal Agents

DevFeed: [Securing Your Newest Customer Channel with Auth for Personal Agents](<https://devfeed.tech/articles/securing-your-newest-customer-channel-with-auth-for-personal-agents-142027.md>)

Original publisher: [Read original article](<https://auth0.com/blog/auth-for-personal-agents-beta-release/>)

Author: Karim Tantawy

Published: 2026-11-18T00:00:00Z

Content type: release

Language: en

Sources: [Auth0 Blog](<https://devfeed.tech/sources/auth0-blog.md>)

Topics: [Auth0](<https://devfeed.tech/topics/auth0.md>), [agentic browsing](<https://devfeed.tech/topics/agentic-browsing.md>), [agent to agent communication](<https://devfeed.tech/topics/agent-to-agent-communication.md>)

Tags: [actions](<https://devfeed.tech/tags/actions.md>), [agent](<https://devfeed.tech/tags/agent.md>), [agents](<https://devfeed.tech/tags/agents.md>), [auth](<https://devfeed.tech/tags/auth.md>), [auth0](<https://devfeed.tech/tags/auth0.md>), [beta](<https://devfeed.tech/tags/beta.md>), [channel](<https://devfeed.tech/tags/channel.md>), [customer](<https://devfeed.tech/tags/customer.md>), [customers](<https://devfeed.tech/tags/customers.md>), [identity](<https://devfeed.tech/tags/identity.md>), [join](<https://devfeed.tech/tags/join.md>), [like](<https://devfeed.tech/tags/like.md>), [muse](<https://devfeed.tech/tags/muse.md>), [personal](<https://devfeed.tech/tags/personal.md>), [personal-agent](<https://devfeed.tech/tags/personal-agent.md>), [personal-agent-protocol](<https://devfeed.tech/tags/personal-agent-protocol.md>), [personal-agents](<https://devfeed.tech/tags/personal-agents.md>), [poppy](<https://devfeed.tech/tags/poppy.md>), [program](<https://devfeed.tech/tags/program.md>), [with](<https://devfeed.tech/tags/with.md>)

### AI overview

Poppy gives personal agents a way to identify themselves and act for customers through scoped, revocable access, avoiding shared passwords and blanket account access. Auth0 says its upcoming beta will support API and web browsing: brands register allowed agents by CIMD URL, customers sign in and consent through standard OAuth, and DPoP-bound tokens plus session checks let services verify and audit the agent separately from the user. The SDK is described as handling discovery metadata and agent-driven web sessions, with CLI setup and diagnostics. A browser playground can test discovery, sign-in, browsing, and conversation against a shop. The beta is coming soon and participation is by selection.

### Source excerpt

Auth0 brings identity to Personal Agents by adding support for the Personal Agent Protocol (Poppy). Verify agents like Muse, tie actions to real customers, and join the Beta Program.

## Updates to our blog policy

DevFeed: [Updates to our blog policy](<https://devfeed.tech/articles/updates-to-our-blog-policy-140175.md>)

Original publisher: [Read original article](<https://opentelemetry.io/blog/2026/blog-policy-update/>)

Author: OpenTelemetry Authors; Docs CC BY

Published: 2026-11-17T18:23:18Z

Content type: news

Language: en

Sources: [OpenTelemetry](<https://devfeed.tech/sources/blog-on-opentelemetry.md>)

Topics: [OpenTelemetry](<https://devfeed.tech/topics/opentelemetry.md>), [agentic browsing](<https://devfeed.tech/topics/agentic-browsing.md>)

Tags: [agents](<https://devfeed.tech/tags/agents.md>), [ai](<https://devfeed.tech/tags/ai.md>), [ai-tools](<https://devfeed.tech/tags/ai-tools.md>), [announcements](<https://devfeed.tech/tags/announcements.md>), [blog](<https://devfeed.tech/tags/blog.md>), [blog-post](<https://devfeed.tech/tags/blog-post.md>), [blog-posts](<https://devfeed.tech/tags/blog-posts.md>), [blogging](<https://devfeed.tech/tags/blogging.md>), [community](<https://devfeed.tech/tags/community.md>), [content](<https://devfeed.tech/tags/content.md>), [content-creators](<https://devfeed.tech/tags/content-creators.md>), [discussion](<https://devfeed.tech/tags/discussion.md>), [documentation](<https://devfeed.tech/tags/documentation.md>), [milestones](<https://devfeed.tech/tags/milestones.md>), [news](<https://devfeed.tech/tags/news.md>), [opentelemetry](<https://devfeed.tech/tags/opentelemetry.md>), [organizational](<https://devfeed.tech/tags/organizational.md>), [policy](<https://devfeed.tech/tags/policy.md>), [releases](<https://devfeed.tech/tags/releases.md>), [updates](<https://devfeed.tech/tags/updates.md>)

### AI overview

OpenTelemetry will keep its blog but restrict submissions to members in good standing of the organization and strongly favor posts about organizational or project announcements, such as releases, milestones, and SIG activity. The stated rationale is a shift in how people find project information: Google referrals have fallen 50% since January, while the site recorded over 15 million page views, nearly a million unique visitors, and 1.4 TB of bandwidth in the last 30 days. The post says rising submission and editorial review demands are less valuable as agents and models use documentation directly, and suggests maintainer effort may be better spent improving documentation. It invites community ideas for surfacing and promoting other OpenTelemetry content.

### Source excerpt

The OpenTelemetry blog is one of our most valuable communication avenues with our community. The web is changing, though, and we need to change with it. Like many websites, we've seen organic search collapse: 50% fewer visitors arrive on opentelemetry.io via Google today than they did in January. This doesn't mean that traffic is down, though. Over the last 30 days, we've had over 15 million page views, nearly a million unique visitors, and 1.4 TB of bandwidth usage.

## Agent Vault gives your agents a signed-in browser for websites without APIs

DevFeed: [Agent Vault gives your agents a signed-in browser for websites without APIs](<https://devfeed.tech/articles/agent-vault-gives-your-agents-a-signed-in-browser-for-websites-without-apis-140649.md>)

Original publisher: [Read original article](<https://workos.com/blog/agent-vault>)

Author: WorkOS

Published: 2026-11-17T00:00:00Z

Content type: release

Language: en

Sources: [WorkOS](<https://devfeed.tech/sources/workos-blog.md>)

Topics: [Agent Browser](<https://devfeed.tech/topics/agent-browser.md>), [agentic browsing](<https://devfeed.tech/topics/agentic-browsing.md>), [human in the loop agent](<https://devfeed.tech/topics/human-in-the-loop-agent.md>)

Tags: [access-tokens](<https://devfeed.tech/tags/access-tokens.md>), [agent](<https://devfeed.tech/tags/agent.md>), [agentic](<https://devfeed.tech/tags/agentic.md>), [agents](<https://devfeed.tech/tags/agents.md>), [ai](<https://devfeed.tech/tags/ai.md>), [ai-agent](<https://devfeed.tech/tags/ai-agent.md>), [api](<https://devfeed.tech/tags/api.md>), [apis](<https://devfeed.tech/tags/apis.md>), [application](<https://devfeed.tech/tags/application.md>), [browser](<https://devfeed.tech/tags/browser.md>), [build](<https://devfeed.tech/tags/build.md>), [connect](<https://devfeed.tech/tags/connect.md>), [hosted](<https://devfeed.tech/tags/hosted.md>), [self-hosted](<https://devfeed.tech/tags/self-hosted.md>), [store](<https://devfeed.tech/tags/store.md>), [users](<https://devfeed.tech/tags/users.md>), [vault](<https://devfeed.tech/tags/vault.md>), [website](<https://devfeed.tech/tags/website.md>), [websites](<https://devfeed.tech/tags/websites.md>), [workos](<https://devfeed.tech/tags/workos.md>)

### AI overview

Agent Vault lets an agent work in a user-authenticated browser on websites without a usable API. Users save logins through a WorkOS widget; Agent Vault handles sign-in and keeps saved passwords out of the agent's prompts and tool responses. Agents connect through MCP to a hosted browser, while users can take over to complete steps such as SSO, CAPTCHA, or passkey checks and return the session. Developers can use the WorkOS-hosted browser or provide one through an existing browser integration; only one browser provider can be chosen per environment. Pipes provides OAuth tokens for supported integrations, while Agent Vault serves browser-only sites, and an application can use both. Access is enabled by the WorkOS team and the product is not generally available.

### Source excerpt

Build agentic products with Agent Vault: store users' website logins and connect agents through a WorkOS-hosted or self-hosted browser. Contact us for access.

## Stop handing over web experiences. Start running a practice

DevFeed: [Stop handing over web experiences. Start running a practice](<https://devfeed.tech/articles/stop-handing-over-web-experiences-start-running-a-practice-136530.md>)

Original publisher: [Read original article](<https://www.webflow.com/blog/stop-handing-over-web-experiences-start-running-a-practice>)

Author: Miles Cox

Published: 2026-11-16T00:00:00Z

Content type: opinion

Language: en

Sources: [Webflow Blog](<https://devfeed.tech/sources/webflow-blog.md>)

Topics: [Website](<https://devfeed.tech/topics/website.md>), [webflow](<https://devfeed.tech/topics/webflow.md>), [agentic browsing](<https://devfeed.tech/topics/agentic-browsing.md>)

Tags: [agencies](<https://devfeed.tech/tags/agencies.md>), [agentic](<https://devfeed.tech/tags/agentic.md>), [agents](<https://devfeed.tech/tags/agents.md>), [ai](<https://devfeed.tech/tags/ai.md>), [ai-agents](<https://devfeed.tech/tags/ai-agents.md>), [ai-models](<https://devfeed.tech/tags/ai-models.md>), [ai-tools](<https://devfeed.tech/tags/ai-tools.md>), [api](<https://devfeed.tech/tags/api.md>), [brand](<https://devfeed.tech/tags/brand.md>), [build](<https://devfeed.tech/tags/build.md>), [business](<https://devfeed.tech/tags/business.md>), [change](<https://devfeed.tech/tags/change.md>), [changes](<https://devfeed.tech/tags/changes.md>), [client](<https://devfeed.tech/tags/client.md>), [clients](<https://devfeed.tech/tags/clients.md>), [code](<https://devfeed.tech/tags/code.md>), [craft](<https://devfeed.tech/tags/craft.md>), [future](<https://devfeed.tech/tags/future.md>), [humans](<https://devfeed.tech/tags/humans.md>), [inspiration](<https://devfeed.tech/tags/inspiration.md>), [web](<https://devfeed.tech/tags/web.md>)

### AI overview

The article argues that agencies should treat website launch as the start of ongoing work: monitor traffic and competitors, test and publish changes quickly, and keep sites current for both people and AI agents. It frames agency value as human taste and distinctive ideas, paired with readable site structure; AI-driven speed alone can produce generic work. On Webflow, marketers can make routine changes using approved components, permissions, and version history, while developers handle harder work. VML's South Bank example reports 44% faster time to market and 36% quicker changes, with changes shipped by the client team in minutes; the article says Webflow is not right for every client. It also suggests checking how major AI tools describe a client's brand and identifying errors as opportunities to update the site. Source by Webflow is described as being in limited research preview.

### Source excerpt

Finished is a flatline. The future belongs to living web experiences, built for humans and the agents that browse for them.

## ChatGPT Wallet lets agents pay without seeing your card number

DevFeed: [ChatGPT Wallet lets agents pay without seeing your card number](<https://devfeed.tech/articles/chatgpt-wallet-lets-agents-pay-without-seeing-your-card-number-132409.md>)

Original publisher: [Read original article](<https://www.helpnetsecurity.com/2026/11/13/openai-chatgpt-wallet/>)

Author: Anamarija Pogorelec

Published: 2026-11-13T16:15:25Z

Content type: news

Language: en

Sources: [Help Net Security](<https://devfeed.tech/sources/help-net-security.md>)

Topics: [digital wallet](<https://devfeed.tech/topics/digital-wallet.md>), [AI Bots](<https://devfeed.tech/topics/ai-bots.md>), [human in the loop agent](<https://devfeed.tech/topics/human-in-the-loop-agent.md>), [agentic browsing](<https://devfeed.tech/topics/agentic-browsing.md>)

Tags: [agentic-ai](<https://devfeed.tech/tags/agentic-ai.md>), [agentic-commerce](<https://devfeed.tech/tags/agentic-commerce.md>), [agents](<https://devfeed.tech/tags/agents.md>), [ai](<https://devfeed.tech/tags/ai.md>), [american-express](<https://devfeed.tech/tags/american-express.md>), [auth](<https://devfeed.tech/tags/auth.md>), [automated](<https://devfeed.tech/tags/automated.md>), [bot](<https://devfeed.tech/tags/bot.md>), [bots](<https://devfeed.tech/tags/bots.md>), [card](<https://devfeed.tech/tags/card.md>), [cdn](<https://devfeed.tech/tags/cdn.md>), [chatgpt](<https://devfeed.tech/tags/chatgpt.md>), [chatgpt-pro](<https://devfeed.tech/tags/chatgpt-pro.md>), [checkout](<https://devfeed.tech/tags/checkout.md>), [digital-wallet](<https://devfeed.tech/tags/digital-wallet.md>), [dots](<https://devfeed.tech/tags/dots.md>), [mastercard](<https://devfeed.tech/tags/mastercard.md>), [news](<https://devfeed.tech/tags/news.md>), [online-payment](<https://devfeed.tech/tags/online-payment.md>), [openai](<https://devfeed.tech/tags/openai.md>), [security](<https://devfeed.tech/tags/security.md>), [visa](<https://devfeed.tech/tags/visa.md>), [wallet](<https://devfeed.tech/tags/wallet.md>), [websites](<https://devfeed.tech/tags/websites.md>)

### AI overview

ChatGPT Wallet lets eligible users approve agent purchases without sharing their card number with the agent or merchant. A user adds a card, reviews the items, total, and selected card for each purchase, then approves with Face ID, a fingerprint, or a similar check. The payment network issues a token with its own security code and expiration date; the charge goes to the underlying card, which retains its rewards and protections for eligible disputes. OpenAI's Commerce Partner Playbook asks merchants to configure CDNs or firewalls to recognize Web Bot Auth and coordinate with payment and risk providers. Existing checkout flows can accept the tokens, but fraud systems may still reject legitimate agent purchases, and sites may block them unintentionally. OpenAI lists collaborations with fraud vendors Forter, Riskified, Sardine, and Signifyd, and payment providers Adyen, Checkout.com, and Stripe. The rollout is limited to US ChatGPT Pro subscribers aged 18 or older with eligible US-issued cards.

### Source excerpt

OpenAI began rolling out ChatGPT Wallet, which lets ChatGPT agents pay for purchases with a stand-in card number, so the user's card details are not shared with the agent or the merchant. OpenAI built it with American Express, Mastercard, and Visa. It runs inside dots, where a user's dot completes the purchase on their behalf. Merchants will have to do most of the adjusting. OpenAI says some websites may block agent purchases without meaning to, ... More -> The post ChatGPT Wallet lets agents pay without seeing your card number appeared first on Help Net Security.

## Introducing ChatGPT Wallet

DevFeed: [Introducing ChatGPT Wallet](<https://devfeed.tech/articles/introducing-chatgpt-wallet-133625.md>)

Original publisher: [Read original article](<https://openai.com/index/introducing-chatgpt-wallet>)

Published: 2026-11-13T15:00:00Z

Content type: release

Language: en

Sources: [OpenAI News](<https://devfeed.tech/sources/openai-news.md>)

Topics: [digital wallet](<https://devfeed.tech/topics/digital-wallet.md>), [human-in-the-loop UX](<https://devfeed.tech/topics/human-in-the-loop-ux.md>), [agentic browsing](<https://devfeed.tech/topics/agentic-browsing.md>)

Tags: [actions](<https://devfeed.tech/tags/actions.md>), [adapting](<https://devfeed.tech/tags/adapting.md>), [adyen](<https://devfeed.tech/tags/adyen.md>), [agent](<https://devfeed.tech/tags/agent.md>), [agentic](<https://devfeed.tech/tags/agentic.md>), [agentic-commerce](<https://devfeed.tech/tags/agentic-commerce.md>), [agents](<https://devfeed.tech/tags/agents.md>), [american-express](<https://devfeed.tech/tags/american-express.md>), [auth](<https://devfeed.tech/tags/auth.md>), [card](<https://devfeed.tech/tags/card.md>), [chatgpt](<https://devfeed.tech/tags/chatgpt.md>), [chatgpt-wallet](<https://devfeed.tech/tags/chatgpt-wallet.md>), [chatgpt-work](<https://devfeed.tech/tags/chatgpt-work.md>), [codex](<https://devfeed.tech/tags/codex.md>), [complete](<https://devfeed.tech/tags/complete.md>), [dots](<https://devfeed.tech/tags/dots.md>), [existing](<https://devfeed.tech/tags/existing.md>), [product](<https://devfeed.tech/tags/product.md>), [user](<https://devfeed.tech/tags/user.md>), [wallet](<https://devfeed.tech/tags/wallet.md>), [work](<https://devfeed.tech/tags/work.md>)

### AI overview

ChatGPT Wallet lets an agent pay with a user's selected card while withholding the card's actual details from the agent and business. Before each purchase, the user reviews the items, total, and card, then approves with a security check such as Face ID or fingerprint; the payment network supplies a token, and the user receives a record of the authorization. Eligible card rewards and dispute protections remain in place, while businesses retain control of pricing, fulfillment, returns, and customer service. The initial rollout is through dots for ChatGPT Pro subscribers aged 18 or older in the US with eligible US-issued cards. Businesses can update verified product details and feeds, allow ChatGPT agents through site security using signed Web Bot Auth requests, and work with payment and risk providers to reduce declines. The article says token payments fit existing checkout flows, though risk systems may still flag legitimate agent purchases. Future delegated purchases with limits and pre-approval are described as an expectation, not a current feature.

### Source excerpt

Introducing ChatGPT Wallet, a secure way for agents in ChatGPT Work, Codex, and dots to complete purchases with user approval and existing card protections.

## Whose Agent Are You? Multi-Layer Fingerprinting and Attribution of Autonomous Web Agents

DevFeed: [Whose Agent Are You? Multi-Layer Fingerprinting and Attribution of Autonomous Web Agents](<https://devfeed.tech/articles/whose-agent-are-you-multi-layer-fingerprinting-and-attribution-of-autonomous-web-agents-131938.md>)

Original publisher: [Read original article](<https://arxiv.org/abs/2606.20910>)

Author: Dayeon Kang, Hyejun Jeong, Jade Sheffey, Pubali Datta, Amir Houmansadr

Published: 2026-11-13T05:00:00Z

Content type: article

Language: en

Sources: [arXiv: Artificial Intelligence](<https://devfeed.tech/sources/arxiv-artificial-intelligence.md>), [arXiv: Cryptography and Security](<https://devfeed.tech/sources/arxiv-cryptography-and-security.md>)

Topics: [AI Bots](<https://devfeed.tech/topics/ai-bots.md>), [agentic browsing](<https://devfeed.tech/topics/agentic-browsing.md>), [TLS (Transport Layer Security)](<https://devfeed.tech/topics/tls.md>), [frontend security](<https://devfeed.tech/topics/frontend-security.md>), [Browser Automation](<https://devfeed.tech/topics/browser-automation.md>)

Tags: [actions](<https://devfeed.tech/tags/actions.md>), [agent](<https://devfeed.tech/tags/agent.md>), [agent-frameworks](<https://devfeed.tech/tags/agent-frameworks.md>), [agents](<https://devfeed.tech/tags/agents.md>), [ai](<https://devfeed.tech/tags/ai.md>), [ai-web](<https://devfeed.tech/tags/ai-web.md>), [architectures](<https://devfeed.tech/tags/architectures.md>), [attribution](<https://devfeed.tech/tags/attribution.md>), [autogen](<https://devfeed.tech/tags/autogen.md>), [autonomous](<https://devfeed.tech/tags/autonomous.md>), [behavior](<https://devfeed.tech/tags/behavior.md>), [bot](<https://devfeed.tech/tags/bot.md>), [browser](<https://devfeed.tech/tags/browser.md>), [browsing](<https://devfeed.tech/tags/browsing.md>), [chatgpt](<https://devfeed.tech/tags/chatgpt.md>), [claude](<https://devfeed.tech/tags/claude.md>), [content](<https://devfeed.tech/tags/content.md>), [content-protection](<https://devfeed.tech/tags/content-protection.md>), [crawlers](<https://devfeed.tech/tags/crawlers.md>), [cs-ai](<https://devfeed.tech/tags/cs-ai.md>), [cs-cr](<https://devfeed.tech/tags/cs-cr.md>), [web](<https://devfeed.tech/tags/web.md>)

### AI overview

The paper presents a multi-layer fingerprinting method to distinguish autonomous web agents from human visitors and traditional crawlers. A logging framework on an instrumented domain combines network characteristics, including TLS and HTTP behavior, with browser interaction patterns. Tested on AutoGen, Browser Use, Claude, Gemini, ChatGPT, and Skyvern, the features reveal differences in request construction, connection setup, and browser actions. A decision tree classifier reportedly achieved 100% accuracy in the study, identifying agent architectures and separating their traffic from human and legacy-crawler baselines. The authors propose cross-layer tracking as a way to support content protection and web security policy enforcement; the supplied abstract does not detail the evaluation conditions or limits of that accuracy result.

### Source excerpt

arXiv:2606.20910v2 Announce Type: replace-cross Abstract: As AI web agents proliferate, combining large language models with autonomous, browser-level control, indiscriminate content scraping by web agents has emerged as a privacy and security challenge. Existing defenses, such as robots.txt and active bot-blocking, are insufficient, as they are widely violated and easily circumvented. In this work, we demonstrate that AI web agents can be effectively distinguished from humans and traditional crawlers using a multi-layer fingerprint based on both network layer characteristics (e.g., TLS, HTTP) and browser interaction behavior. We implement this mechanism as a programmatic logging framework that can be deployed on a live, instrumented domain. By analyzing six prominent agent frameworks (AutoGen, Browser Use, Claude, Gemini, ChatGPT, and Skyvern), we uncover latent structural differences in how these systems assemble HTTP requests, establish TLS/HTTP connections, and execute autonomous browser actions. Feeding these multi-layer features into a decision tree classifier, our framework achieves high-fidelity identification (100% accuracy), successfully isolating distinct agent architectures and differentiating agent traffic from both human browsing baselines and legacy crawlers. Our findings demonstrate that cross-layer agent tracking provides a robust approach to content protection and web security policy enforcement.

## Known By Their Actions: Fingerprinting LLM Browser Agents via UI Traces

DevFeed: [Known By Their Actions: Fingerprinting LLM Browser Agents via UI Traces](<https://devfeed.tech/articles/known-by-their-actions-fingerprinting-llm-browser-agents-via-ui-traces-128107.md>)

Original publisher: [Read original article](<https://arxiv.org/abs/2605.14786>)

Author: William Lugoloobi, Samuele Marro, Jabez Magomere, Joss Wright, Chris Russell

Published: 2026-11-11T05:00:00Z

Content type: news

Language: en

Sources: [arXiv: Artificial Intelligence](<https://devfeed.tech/sources/arxiv-artificial-intelligence.md>), [arXiv: Cryptography and Security](<https://devfeed.tech/sources/arxiv-cryptography-and-security.md>), [arXiv: Machine Learning](<https://devfeed.tech/sources/arxiv-machine-learning.md>)

Topics: [LLM security](<https://devfeed.tech/topics/llm-security.md>), [Large Language Model](<https://devfeed.tech/topics/llm.md>), [agentic browsing](<https://devfeed.tech/topics/agentic-browsing.md>)

Tags: [agents](<https://devfeed.tech/tags/agents.md>), [attacks](<https://devfeed.tech/tags/attacks.md>), [browser](<https://devfeed.tech/tags/browser.md>), [browser-agents](<https://devfeed.tech/tags/browser-agents.md>), [classifiers](<https://devfeed.tech/tags/classifiers.md>), [code](<https://devfeed.tech/tags/code.md>), [cs-ai](<https://devfeed.tech/tags/cs-ai.md>), [cs-cr](<https://devfeed.tech/tags/cs-cr.md>), [cs-hc](<https://devfeed.tech/tags/cs-hc.md>), [cs-lg](<https://devfeed.tech/tags/cs-lg.md>), [environments](<https://devfeed.tech/tags/environments.md>), [events](<https://devfeed.tech/tags/events.md>), [interaction](<https://devfeed.tech/tags/interaction.md>), [javascript](<https://devfeed.tech/tags/javascript.md>), [llm](<https://devfeed.tech/tags/llm.md>), [llms](<https://devfeed.tech/tags/llms.md>), [model](<https://devfeed.tech/tags/model.md>), [performance](<https://devfeed.tech/tags/performance.md>), [protection](<https://devfeed.tech/tags/protection.md>), [security](<https://devfeed.tech/tags/security.md>), [shopping](<https://devfeed.tech/tags/shopping.md>), [targeted-attacks](<https://devfeed.tech/tags/targeted-attacks.md>), [traces](<https://devfeed.tech/tags/traces.md>), [ui](<https://devfeed.tech/tags/ui.md>)

### AI overview

A passive JavaScript tracker can identify which of 14 frontier LLMs powers a browser agent from its UI actions and interaction timings, reaching up to 96% F1 within a fixed harness across four information retrieval and shopping environments. Fingerprints transfer poorly across harnesses, though pooling traces across harnesses yields 0.74-0.81 Macro-F1. Classifiers perform well with roughly one third of the available traces, and accuracy improves as more events are observed. Randomized delays substantially reduce performance, but are not robust protection: retraining on delayed traces largely restores it. The authors release code and a labeled trace corpus.

### Source excerpt

arXiv:2605.14786v2 Announce Type: replace-cross Abstract: As LLM-based agents increasingly browse the web on users' behalf, a natural question arises: can websites passively identify which underlying model powers an agent? Doing so would represent a significant security risk, enabling targeted attacks tailored to known model vulnerabilities. Across 14 frontier LLMs and four web environments spanning information retrieval and shopping tasks, we show that an agent's actions and interaction timings, captured via a passive JavaScript tracker, are sufficient to identify the underlying model with up to 96\% F1 within a fixed harness. Although fingerprints are harness-specific, with cross-harness transfer falling to near chance, pooling traces across harnesses recovers model identification at 0.74 to 0.81 Macro-F1. We further show that strong classifiers can be trained from roughly one third of the available traces, and that identification improves steadily as more events are observed. Injecting randomised timing delays between actions substantially degrades classifier performance, but does not provide robust protection: a classifier retrained on delayed traces largely recovers performance. We release our code and a labelled corpus of agent traces at https://github.com/KabakaWilliam/known_actions.

## Harness Matters: Measuring and Dissecting Security Resilience in Web Agents

DevFeed: [Harness Matters: Measuring and Dissecting Security Resilience in Web Agents](<https://devfeed.tech/articles/harness-matters-measuring-and-dissecting-security-resilience-in-web-agents-126209.md>)

Original publisher: [Read original article](<https://arxiv.org/abs/2611.11971>)

Author: Xinyi Wu, Jiagui Chen, JiaYi Dong, Geng Hong, Xudong Pan, Jiarun Dai, Mingxuan Liu, Min Yang

Published: 2026-11-10T05:00:00Z

Content type: article

Language: en

Sources: [arXiv: Artificial Intelligence](<https://devfeed.tech/sources/arxiv-artificial-intelligence.md>), [arXiv: Cryptography and Security](<https://devfeed.tech/sources/arxiv-cryptography-and-security.md>)

Topics: [Security](<https://devfeed.tech/topics/security.md>), [Agent Harness](<https://devfeed.tech/topics/agent-harness.md>), [prompt injection](<https://devfeed.tech/topics/prompt-injection.md>), [agentic browsing](<https://devfeed.tech/topics/agentic-browsing.md>), [Resilience](<https://devfeed.tech/topics/resilience.md>)

Tags: [agent-harnesses](<https://devfeed.tech/tags/agent-harnesses.md>), [agent-security](<https://devfeed.tech/tags/agent-security.md>), [agents](<https://devfeed.tech/tags/agents.md>), [cs-ai](<https://devfeed.tech/tags/cs-ai.md>), [cs-cr](<https://devfeed.tech/tags/cs-cr.md>), [evaluation](<https://devfeed.tech/tags/evaluation.md>), [harness](<https://devfeed.tech/tags/harness.md>), [intrinsic](<https://devfeed.tech/tags/intrinsic.md>), [prompt-injection](<https://devfeed.tech/tags/prompt-injection.md>), [security](<https://devfeed.tech/tags/security.md>), [web](<https://devfeed.tech/tags/web.md>), [web-design](<https://devfeed.tech/tags/web-design.md>), [workflow](<https://devfeed.tech/tags/workflow.md>)

### AI overview

Web-agent security depends on both the execution harness and the backend model. The study describes harnesses as four-stage workflows--goal interpretation, webpage perception, reasoning and planning, and action execution--and introduces WebTrap Farm to evaluate malicious user prompts, prompt injection, and deceptive web design. Paired usability calibration tasks are used to distinguish defensive behavior from task-capability limits. The evaluation finds vulnerabilities to typical attacks, including unsafe behavior without explicit attack content, and reports that both harness and model significantly affect security outcomes. The abstract also says the study locates defensive behavior across workflow stages and identifies more secure design mechanisms, though it does not specify those mechanisms.

### Source excerpt

arXiv:2611.11971v1 Announce Type: cross Abstract: Web agents extend LLMs from text generation to real-world web interaction by relying on execution harnesses that interpret user goals, perceive webpages, plan actions, and ground model decisions into browser operations. While prior work has recognized that agent security depends on more than the backend model alone, how harness design quantitatively shapes web-agent security remains insufficiently measured. To systematically study web-agent harness security, we first characterize web-agent harnesses as a four-stage execution workflow covering goal interpretation, webpage perception, reasoning and planning, and action execution. Based on this abstraction, we build WebTrap Farm, a systematic and extensible evaluation framework covering three representative attack categories: malicious user prompts, malicious prompt injection, and deceptive web design. To avoid conflating security with task capability, we use paired usability calibration tasks to separate genuine defensive behavior from capability limitations. Our evaluation shows that existing web-agent harnesses remain vulnerable to typical attacks, with some unsafe behaviors occurring even without explicit attack content, revealing harness-intrinsic security failures (RQ1). Statistical analysis further shows that both harness and backend model significantly affect security outcomes, indicating that deployed-agent security is jointly shaped by the harness and backend model (RQ2). Finally, we localize where defensive behavior emerges across workflow stages and identify more secure design mechanisms, providing actionable guidance for building more robust web agents (RQ3). These findings highlight the web-agent harness as an important security boundary and motivate moving beyond model-centric defenses toward harness-level security design.

## ScrapeBench: Evaluating Legal Compliance of AI Agents in Website Scraping

DevFeed: [ScrapeBench: Evaluating Legal Compliance of AI Agents in Website Scraping](<https://devfeed.tech/articles/scrapebench-evaluating-legal-compliance-of-ai-agents-in-website-scraping-126100.md>)

Original publisher: [Read original article](<https://arxiv.org/abs/2611.11306>)

Author: Joseph Marvin Imperial, Daniel Slate, Noam Kolt

Published: 2026-11-10T05:00:00Z

Content type: article

Language: en

Sources: [arXiv: Artificial Intelligence](<https://devfeed.tech/sources/arxiv-artificial-intelligence.md>)

Topics: [agentic browsing](<https://devfeed.tech/topics/agentic-browsing.md>), [Human-AI evaluation](<https://devfeed.tech/topics/human-ai-evaluation.md>), [audit](<https://devfeed.tech/topics/audit.md>)

Tags: [ai-agents](<https://devfeed.tech/tags/ai-agents.md>), [benchmark](<https://devfeed.tech/tags/benchmark.md>), [compliance](<https://devfeed.tech/tags/compliance.md>), [cs-ai](<https://devfeed.tech/tags/cs-ai.md>), [cs-cl](<https://devfeed.tech/tags/cs-cl.md>), [legal](<https://devfeed.tech/tags/legal.md>), [scraping](<https://devfeed.tech/tags/scraping.md>), [website](<https://devfeed.tech/tags/website.md>)

### AI overview

ScrapeBench evaluates whether AI agents follow website scraping policies in terms of service and robots.txt. Across 1,626 sites that explicitly prohibit scraping, most tested agents complied at rates below 5%, and checking the policies did not substantially improve compliance. On 269 sites that did not prohibit scraping, agents sometimes refused; Claude Opus 4.6 had a reported over-refusal rate of 98.5%. Human participants showed compliance comparable to the agents but over-refused more often. These findings are limited to the benchmark's samples and experiments.

### Source excerpt

arXiv:2611.11306v1 Announce Type: cross Abstract: 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 (< 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.

## EcoGEO: Trajectory-Aware Evidence Ecosystems for Web-Enabled LLM Search Agents

DevFeed: [EcoGEO: Trajectory-Aware Evidence Ecosystems for Web-Enabled LLM Search Agents](<https://devfeed.tech/articles/ecogeo-trajectory-aware-evidence-ecosystems-for-web-enabled-llm-search-agents-120026.md>)

Original publisher: [Read original article](<https://arxiv.org/abs/2605.12887>)

Author: Hengwei Ye, Jiasheng Mao, Zhenhan Guan, Zheng Tian

Published: 2026-11-06T05:00:00Z

Content type: article

Language: en

Sources: [arXiv: Artificial Intelligence](<https://devfeed.tech/sources/arxiv-artificial-intelligence.md>)

Topics: [search-engines](<https://devfeed.tech/topics/search-engines.md>), [agentic browsing](<https://devfeed.tech/topics/agentic-browsing.md>), [Search engine optimization (SEO)](<https://devfeed.tech/topics/seo.md>)

Tags: [agents](<https://devfeed.tech/tags/agents.md>), [browsing](<https://devfeed.tech/tags/browsing.md>), [cs-ai](<https://devfeed.tech/tags/cs-ai.md>), [cs-ir](<https://devfeed.tech/tags/cs-ir.md>), [ecosystem](<https://devfeed.tech/tags/ecosystem.md>), [ecosystems](<https://devfeed.tech/tags/ecosystems.md>), [generative](<https://devfeed.tech/tags/generative.md>), [generative-engine-optimization](<https://devfeed.tech/tags/generative-engine-optimization.md>), [geo](<https://devfeed.tech/tags/geo.md>), [llm](<https://devfeed.tech/tags/llm.md>), [llm-agents](<https://devfeed.tech/tags/llm-agents.md>), [multiple](<https://devfeed.tech/tags/multiple.md>), [navigation](<https://devfeed.tech/tags/navigation.md>), [optimization](<https://devfeed.tech/tags/optimization.md>), [organization](<https://devfeed.tech/tags/organization.md>), [pages](<https://devfeed.tech/tags/pages.md>), [process](<https://devfeed.tech/tags/process.md>), [product](<https://devfeed.tech/tags/product.md>), [queries](<https://devfeed.tech/tags/queries.md>), [query](<https://devfeed.tech/tags/query.md>), [search](<https://devfeed.tech/tags/search.md>), [web](<https://devfeed.tech/tags/web.md>)

### AI overview

EcoGEO treats generative engine optimization as shaping the broader evidence environment a web-enabled search agent encounters, rather than optimizing a single webpage. Its TRACE method builds a controlled set of pages for a recommendation query and fictional target product: an agent-facing navigation page plus varied support pages linked internally and aligned through shared terminology and consistent product attributes. On OPR-Bench, TRACE consistently outperformed page-level GEO baselines in recommendation. It improved initial access to the target and internal-link browsing, while effects on follow-up searches varied by model backbone and comparison method. The abstract presents these results as evidence that page organization can shape evidence acquisition; it does not establish uniform gains across follow-up searches.

### Source excerpt

arXiv:2605.12887v3 Announce Type: replace-cross Abstract: Web-enabled LLM agents are changing how online information influences search outcomes. Existing Generative Engine Optimization (GEO) studies mainly focus on individual webpages. However, agentic web search is not a single-document setting: an agent may issue queries, crawl pages, follow links, reformulate searches, and synthesize evidence across multiple browsing steps. Influence therefore depends not only on page content, but also on how pages are organized, connected, and encountered along the agent's browsing trajectory. We study this shift through Ecosystem Generative Engine Optimization (EcoGEO), which treats GEO as an environment-level influence problem for web-enabled LLM agents. To instantiate this perspective, we propose TRACE, a Trajectory-Aware Coordinated Evidence Ecosystem. Given a recommendation query and a fictional target product, our method builds a controlled evidence environment that coordinates an agent-facing navigation entry page with heterogeneous support pages. These pages use shared terminology, internal links, and consistent product attributes to introduce, verify, and reinforce the target product. We evaluate our method on OPR-Bench, a benchmark for open-ended product recommendation. Experiments show that it consistently outperforms page-level GEO baselines in Recommendation. Trajectory-level metrics show improved initial target access and additional internal-link browsing, while follow-up-search effects vary across backbones and comparison methods. Together with the multi-page baseline and ablations, these findings support the role of evidence organization in shaping the agent's evidence-acquisition process. Overall, our findings support an ecosystem research paradigm for GEO, where web-enabled LLM agents are studied in relation to the broader evidence environments that guide search, browsing, and answer synthesis.

## Friday Links #46: The Latest in JavaScript and Web Development

DevFeed: [Friday Links #46: The Latest in JavaScript and Web Development](<https://devfeed.tech/articles/friday-links-46-the-latest-in-javascript-and-web-development-121775.md>)

Original publisher: [Read original article](<https://jsdev.space/friday/friday-46/>)

Author: anton liberant

Published: 2026-11-06T00:00:00Z

Content type: news

Language: en

Sources: [JavaScript Development](<https://devfeed.tech/sources/javascript-development.md>)

Topics: [Web Development](<https://devfeed.tech/topics/web-development.md>), [Front end](<https://devfeed.tech/topics/frontend.md>), [Node.js](<https://devfeed.tech/topics/node-js.md>), [React](<https://devfeed.tech/topics/react.md>), [TypeScript](<https://devfeed.tech/topics/typescript.md>), [Google Chrome](<https://devfeed.tech/topics/google-chrome.md>), [Nest](<https://devfeed.tech/topics/nestjs.md>), [Svelte](<https://devfeed.tech/topics/svelte.md>), [Cloudflare Workers](<https://devfeed.tech/topics/cloudflare-workers.md>), [agentic browsing](<https://devfeed.tech/topics/agentic-browsing.md>), [web animation](<https://devfeed.tech/topics/web-animation.md>)

Tags: [agent](<https://devfeed.tech/tags/agent.md>), [agentic](<https://devfeed.tech/tags/agentic.md>), [agentic-workflows](<https://devfeed.tech/tags/agentic-workflows.md>), [agents](<https://devfeed.tech/tags/agents.md>), [ai](<https://devfeed.tech/tags/ai.md>), [ai-agent](<https://devfeed.tech/tags/ai-agent.md>), [ai-assisted](<https://devfeed.tech/tags/ai-assisted.md>), [animations](<https://devfeed.tech/tags/animations.md>), [browser](<https://devfeed.tech/tags/browser.md>), [developer-tools](<https://devfeed.tech/tags/developer-tools.md>), [friday](<https://devfeed.tech/tags/friday.md>), [javascript](<https://devfeed.tech/tags/javascript.md>), [news](<https://devfeed.tech/tags/news.md>), [node](<https://devfeed.tech/tags/node.md>), [node-js](<https://devfeed.tech/tags/node-js.md>), [react](<https://devfeed.tech/tags/react.md>), [tools](<https://devfeed.tech/tags/tools.md>), [typescript](<https://devfeed.tech/tags/typescript.md>), [updates](<https://devfeed.tech/tags/updates.md>), [web-development](<https://devfeed.tech/tags/web-development.md>)

### AI overview

The roundup reports a mix of runtime, framework, tooling, and platform updates. Node.js 26.13.0 is now LTS, while Node.js 27 is available only as an alpha for early testing. It highlights security fixes in Next.js 16.4.5 and express-fileupload 1.6, plus performance and tooling changes such as Rspack's reported 5-20% faster builds, Vitest's affected-test reruns, and pnpm's global peer dependency checks. Other notable changes include Preact's React Compiler compatibility, NestJS's native `node:http` adapter, and Leaflet 2.0 beta's built-in TypeScript types. The news section says Discord is rebuilding its apps in Rust, with no public release date announced, and describes new Chrome APIs and GitHub agentic workflows reaching general availability. The excerpt is a broad roundup rather than detailed coverage of each release.

### Source excerpt

Explore Friday Links #46: the latest JavaScript, TypeScript, React, Node.js, browser updates, developer tools, and web development news.

## Beyond the Landing Page: Website Blocking, Tracker Visibility, and Human Alignment in Automated Web Crawls

DevFeed: [Beyond the Landing Page: Website Blocking, Tracker Visibility, and Human Alignment in Automated Web Crawls](<https://devfeed.tech/articles/beyond-the-landing-page-website-blocking-tracker-visibility-and-human-alignment-in-automated-web-crawls-118704.md>)

Original publisher: [Read original article](<https://arxiv.org/abs/2611.06197>)

Author: A H M Nazmus Sakib, Mahsin Bin Akram, Raveen Wijewickrama, Igor Bilogrevic, Murtuza Jadliwala

Published: 2026-11-05T05:00:00Z

Content type: article

Language: en

Sources: [arXiv: Cryptography and Security](<https://devfeed.tech/sources/arxiv-cryptography-and-security.md>)

Topics: [Web Scraping](<https://devfeed.tech/topics/web-scraping.md>), [AI Bots](<https://devfeed.tech/topics/ai-bots.md>), [Browser Automation](<https://devfeed.tech/topics/browser-automation.md>), [agentic browsing](<https://devfeed.tech/topics/agentic-browsing.md>)

Tags: [1-5](<https://devfeed.tech/tags/1-5.md>), [1-7](<https://devfeed.tech/tags/1-7.md>), [abstract](<https://devfeed.tech/tags/abstract.md>), [agentic](<https://devfeed.tech/tags/agentic.md>), [agents](<https://devfeed.tech/tags/agents.md>), [alignment](<https://devfeed.tech/tags/alignment.md>), [arxiv](<https://devfeed.tech/tags/arxiv.md>), [automated](<https://devfeed.tech/tags/automated.md>), [automation](<https://devfeed.tech/tags/automation.md>), [blocking](<https://devfeed.tech/tags/blocking.md>), [browser](<https://devfeed.tech/tags/browser.md>), [browser-automation](<https://devfeed.tech/tags/browser-automation.md>), [browsing](<https://devfeed.tech/tags/browsing.md>), [can](<https://devfeed.tech/tags/can.md>), [compare](<https://devfeed.tech/tags/compare.md>), [consent](<https://devfeed.tech/tags/consent.md>), [cs-cr](<https://devfeed.tech/tags/cs-cr.md>), [tracker](<https://devfeed.tech/tags/tracker.md>), [visibility](<https://devfeed.tech/tags/visibility.md>), [web](<https://devfeed.tech/tags/web.md>), [website](<https://devfeed.tech/tags/website.md>)

### AI overview

Across 2,000 popular websites, standard browser automation faced more denials beyond landing pages, while a detection-resistant framework largely prevented that increase without recording more tracking. On the same URLs, agentic crawlers recorded tracking from 1.5-1.7 times as many third-party organizations as non-agentic crawlers, especially on consent-managed pages. They captured 82-87% of tracker organizations seen by 26 participants, versus 58-68% for non-agentic crawlers, but about half of agentic crawlers' observations were absent from participant visits. The findings show that limiting website interventions, increasing tracker visibility, and matching human observations are distinct measurement goals; crawlers should be selected and validated accordingly.

### Source excerpt

arXiv:2611.06197v1 Announce Type: new Abstract: Web tracking studies largely rely on automated crawlers, yet their measurements can differ from what people encounter while browsing. Conventional crawlers follow predefined navigation rules, whereas LLM-driven agents decide how to navigate and interact with websites, potentially changing the tracking they encounter. We compare agentic and non-agentic crawlers to examine how websites respond to them, what tracking they record, and how closely their observations match human browsing. Across 2,000 popular websites, we evaluate different navigation strategies, automation frameworks, and stealth configurations, alongside browsing data from 26 participants. We find that website denials increase beyond landing pages for standard browser automation, while a detection-resistant automation framework largely prevents this escalation without increasing recorded tracking. On the same URLs, agentic crawlers record tracking activity from 1.5-1.7 times as many third-party organizations, particularly on consent-managed pages. Agentic crawlers recover 82-87% of tracker organizations observed by participants, compared with 58-68% for non-agentic crawlers, yet roughly half of their own observations are absent from human participant visits. These findings show that avoiding website interventions, increasing tracking visibility, and achieving closer agreement with human observations do not necessarily go hand in hand, highlighting the need to select and validate crawlers according to the measurement objective.

## Securing Web Agents with the Compile-Then-Execute Paradigm

DevFeed: [Securing Web Agents with the Compile-Then-Execute Paradigm](<https://devfeed.tech/articles/securing-web-agents-with-the-compile-then-execute-paradigm-118697.md>)

Original publisher: [Read original article](<https://arxiv.org/abs/2611.06039>)

Author: Annabella Chow, Jinhao Zhu, Sylvie Venuto, Abdulaziz Alharbi, Basel Alomair, Raluca Popa, David Wagner, Julien Piet

Published: 2026-11-05T05:00:00Z

Content type: article

Language: en

Sources: [arXiv: Cryptography and Security](<https://devfeed.tech/sources/arxiv-cryptography-and-security.md>)

Topics: [prompt injection](<https://devfeed.tech/topics/prompt-injection.md>), [agentic browsing](<https://devfeed.tech/topics/agentic-browsing.md>), [Agent Browser](<https://devfeed.tech/topics/agent-browser.md>), [Web](<https://devfeed.tech/topics/web.md>)

Tags: [agent](<https://devfeed.tech/tags/agent.md>), [agents](<https://devfeed.tech/tags/agents.md>), [api](<https://devfeed.tech/tags/api.md>), [benchmark](<https://devfeed.tech/tags/benchmark.md>), [compilation](<https://devfeed.tech/tags/compilation.md>), [compiler](<https://devfeed.tech/tags/compiler.md>), [cs-cr](<https://devfeed.tech/tags/cs-cr.md>), [evaluation](<https://devfeed.tech/tags/evaluation.md>), [hijacking](<https://devfeed.tech/tags/hijacking.md>), [planning](<https://devfeed.tech/tags/planning.md>), [prompt-injection](<https://devfeed.tech/tags/prompt-injection.md>), [python](<https://devfeed.tech/tags/python.md>), [runtime](<https://devfeed.tech/tags/runtime.md>), [security](<https://devfeed.tech/tags/security.md>), [web](<https://devfeed.tech/tags/web.md>), [website](<https://devfeed.tech/tags/website.md>)

### AI overview

Baya aims to prevent webpage prompt injections from hijacking a web agent's control flow by compiling the user's request into a Python program before accessing the live site. The planning model does not see live content; the compiled program processes it through a Trusted Website API. Untrusted content can supply values or choose among precompiled branches, but cannot redefine the task or prompt the model to create new website operations at runtime. Across 581 valid WebArena Verified tasks, Baya reported 77.8% aggregate utility, compared with 66.7% for API-enabled ReAct and 55.1% for browser-only ReAct. On WASP, its reported attack success rate was 0%, versus 16.2% and 25.5%, respectively. These are reported benchmark results; the supplied text does not describe evaluation limitations.

### Source excerpt

arXiv:2611.06039v1 Announce Type: new Abstract: Web agents commonly follow the ReAct paradigm, selecting each action from runtime observations that inherently mix trusted and untrusted content. This design makes their control flow vulnerable to prompt injections embedded in webpages. We introduce Baya, a compile-then-execute web agent designed to prevent prompt-injection-driven control-flow hijacking by design. Before accessing the live website, Baya compiles the user's request into an executable Python program that calls a Trusted Website API. The planning model does not observe live website content; only the compiled program processes it. Untrusted data may determine program values or select among branches defined during compilation, but it cannot redefine the user task or cause the model to synthesize new website operations at runtime. The Trusted Website API gives the compiler a unified view of the website's action space, enabling ahead-of-time planning without the stepwise action discovery required by reactive agents. Our evaluation across 581 valid WebArena Verified tasks shows that Baya achieves 77.8% aggregate utility, outperforming browser-only ReAct (55.1%) and API-enabled ReAct (66.7%). On WASP, a security benchmark for web agents, Baya achieves 0% attack success rate, compared with 16.2% for API-enabled ReAct and 25.5% for browser-only ReAct.

## Citrix expands oversight of autonomous browsing and AI agent activity

DevFeed: [Citrix expands oversight of autonomous browsing and AI agent activity](<https://devfeed.tech/articles/citrix-expands-oversight-of-autonomous-browsing-and-ai-agent-activity-116586.md>)

Original publisher: [Read original article](<https://www.helpnetsecurity.com/2026/11/04/citrix-ai-governance-capabilities/>)

Author: Industry News

Published: 2026-11-04T14:13:48Z

Content type: news

Language: en

Sources: [Help Net Security](<https://devfeed.tech/sources/help-net-security.md>)

Topics: [ai-governance](<https://devfeed.tech/topics/ai-governance.md>), [AI Agent](<https://devfeed.tech/topics/ai-agent.md>), [agentic browsing](<https://devfeed.tech/topics/agentic-browsing.md>)

Tags: [adoption](<https://devfeed.tech/tags/adoption.md>), [agent](<https://devfeed.tech/tags/agent.md>), [agents](<https://devfeed.tech/tags/agents.md>), [ai](<https://devfeed.tech/tags/ai.md>), [ai-agent](<https://devfeed.tech/tags/ai-agent.md>), [ai-agents](<https://devfeed.tech/tags/ai-agents.md>), [ai-governance](<https://devfeed.tech/tags/ai-governance.md>), [ai-powered](<https://devfeed.tech/tags/ai-powered.md>), [autonomous](<https://devfeed.tech/tags/autonomous.md>), [browsing](<https://devfeed.tech/tags/browsing.md>), [capabilities](<https://devfeed.tech/tags/capabilities.md>), [citrix](<https://devfeed.tech/tags/citrix.md>), [deploy](<https://devfeed.tech/tags/deploy.md>), [enterprise](<https://devfeed.tech/tags/enterprise.md>), [flex](<https://devfeed.tech/tags/flex.md>), [flexible](<https://devfeed.tech/tags/flexible.md>), [governance](<https://devfeed.tech/tags/governance.md>), [industry-news](<https://devfeed.tech/tags/industry-news.md>), [organizations](<https://devfeed.tech/tags/organizations.md>), [oversight](<https://devfeed.tech/tags/oversight.md>), [platform](<https://devfeed.tech/tags/platform.md>)

### AI overview

Citrix describes new controls for deploying and governing AI agents across enterprise systems. SecurSpaces provides centrally managed, isolated workspaces where agents receive only task-required resources; credential protections keep sensitive application credentials outside the execution environment. Administrators can monitor activity, access, security events and AI consumption. SecurAccess with Chrome Enterprise is planned to detect automated browser activity, with options to monitor, notify, record sessions or block it. SecurAccess ZTNA is planned to restrict agents to task-specific tools and actions through Model Context Protocol, while Session Insights will flag potentially risky moments and preserve an audit record. Citrix also introduces Platform Credits, a shared consumption pool intended to let customers adjust resources across its portfolio as needs change. The Knowledge Worker Agent is planned for open-source release under Apache 2.0.

### Source excerpt

Citrix has announced new capabilities across Citrix SecurSpaces, Citrix SecurAccess and Citrix Platform Flex designed to help organizations securely deploy AI agents, govern AI-powered work and scale adoption across the enterprise. These updates include an expansion of Citrix SecurSpaces to a wider set of personas; new AI governance capabilities for Citrix SecurAccess; and a new credit-based model offering Citrix customers a more flexible way to consume Citrix solutions and services. As AI evolves from conversational ... More -> The post Citrix expands oversight of autonomous browsing and AI agent activity appeared first on Help Net Security.

## Agents act, LLMs recommend, and people browse. Microsoft Clarity connects all three journeys.

DevFeed: [Agents act, LLMs recommend, and people browse. Microsoft Clarity connects all three journeys.](<https://devfeed.tech/articles/agents-act-llms-recommend-and-people-browse-microsoft-clarity-connects-all-three-journeys-117698.md>)

Original publisher: [Read original article](<https://clarity.microsoft.com/blog/clarity-connects-agents-llms-people/>)

Author: Nabil Naghdy

Published: 2026-11-04T14:00:00Z

Content type: release

Language: en

Sources: [Microsoft Clarity](<https://devfeed.tech/sources/microsoft-clarity-blog.md>)

Topics: [AI Bots](<https://devfeed.tech/topics/ai-bots.md>), [Agentic AI](<https://devfeed.tech/topics/what-is-agentic-ai.md>), [agentic browsing](<https://devfeed.tech/topics/agentic-browsing.md>), [User Experience](<https://devfeed.tech/topics/user-experience.md>), [Shopify](<https://devfeed.tech/topics/shopify.md>)

Tags: [2](<https://devfeed.tech/tags/2.md>), [2026](<https://devfeed.tech/tags/2026.md>), [3](<https://devfeed.tech/tags/3.md>), [about](<https://devfeed.tech/tags/about.md>), [ads](<https://devfeed.tech/tags/ads.md>), [advertising](<https://devfeed.tech/tags/advertising.md>), [agent](<https://devfeed.tech/tags/agent.md>), [agent-experience](<https://devfeed.tech/tags/agent-experience.md>), [agents](<https://devfeed.tech/tags/agents.md>), [ai](<https://devfeed.tech/tags/ai.md>), [ai-agents](<https://devfeed.tech/tags/ai-agents.md>), [all](<https://devfeed.tech/tags/all.md>), [blogs](<https://devfeed.tech/tags/blogs.md>), [customers](<https://devfeed.tech/tags/customers.md>), [data](<https://devfeed.tech/tags/data.md>), [faster](<https://devfeed.tech/tags/faster.md>), [help](<https://devfeed.tech/tags/help.md>), [microsoft](<https://devfeed.tech/tags/microsoft.md>), [people](<https://devfeed.tech/tags/people.md>), [site](<https://devfeed.tech/tags/site.md>), [understand](<https://devfeed.tech/tags/understand.md>)

### AI overview

Microsoft Clarity introduces three capabilities for changing customer journeys: Agent Activity tracks supported AI agents' visits and actions, while simulations test task completion, duration, interactions, and potential trouble spots. Simulations start with commerce and travel. Brand Agents can surface eligible WooCommerce and Shopify products in Microsoft Copilot, Bing, MSN, and other Microsoft experiences; shoppers complete purchases on merchants' sites. For local businesses, agents can answer service questions and connect with supported booking and CRM tools to capture leads or schedule appointments. Ask Clarity analyzes website behavior data, including recordings, heatmaps, funnels, and events, and links findings and recommendations to session evidence. It is in closed beta for retailers; AI Visibility and advertising data are planned additions. The article reports that weekly agent sessions nearly doubled to almost 2 million in three weeks, based on a Clarity analysis of 5.23 million sessions across 230,511 projects.

### Source excerpt

See what AI agents do on your site, get answers from your data faster, and help customers find you across The post Agents act, LLMs recommend, and people browse. Microsoft Clarity connects all three journeys. appeared first on Understand your customers | Microsoft Clarity Blog.

## Making your site visible to LLMs and AI agents: 6 techniques that work, 8 that don't

DevFeed: [Making your site visible to LLMs and AI agents: 6 techniques that work, 8 that don't](<https://devfeed.tech/articles/making-your-site-visible-to-llms-and-ai-agents-6-techniques-that-work-8-that-don-t-118980.md>)

Original publisher: [Read original article](<https://evilmartians.com/chronicles/how-to-make-your-website-visible-to-llms>)

Author: Travis Turner (richardturner@evilmartians.com)

Published: 2026-11-04T00:00:00Z

Content type: article

Language: en

Sources: [Evil Martians](<https://devfeed.tech/sources/evil-martians.md>)

Topics: [Search engine optimization (SEO)](<https://devfeed.tech/topics/seo.md>), [agentic browsing](<https://devfeed.tech/topics/agentic-browsing.md>), [Artificial Intelligence](<https://devfeed.tech/topics/ai.md>), [Claude Code](<https://devfeed.tech/topics/claude-code.md>), [static-site-generator](<https://devfeed.tech/topics/static-site-generator.md>), [AI Chat](<https://devfeed.tech/topics/ai-chat.md>)

Tags: [accessibility](<https://devfeed.tech/tags/accessibility.md>), [adobe](<https://devfeed.tech/tags/adobe.md>), [agent-readiness](<https://devfeed.tech/tags/agent-readiness.md>), [agent-skill](<https://devfeed.tech/tags/agent-skill.md>), [agent-skills](<https://devfeed.tech/tags/agent-skills.md>), [agentic](<https://devfeed.tech/tags/agentic.md>), [agentic-browsing](<https://devfeed.tech/tags/agentic-browsing.md>), [agentic-commerce](<https://devfeed.tech/tags/agentic-commerce.md>), [ai](<https://devfeed.tech/tags/ai.md>), [ai-agent](<https://devfeed.tech/tags/ai-agent.md>), [ai-agents](<https://devfeed.tech/tags/ai-agents.md>), [ai-answers](<https://devfeed.tech/tags/ai-answers.md>), [ai-assistant](<https://devfeed.tech/tags/ai-assistant.md>), [ai-assistants](<https://devfeed.tech/tags/ai-assistants.md>), [ai-bots](<https://devfeed.tech/tags/ai-bots.md>), [chatgpt](<https://devfeed.tech/tags/chatgpt.md>), [claude](<https://devfeed.tech/tags/claude.md>), [dx](<https://devfeed.tech/tags/dx.md>), [frontend-real-time](<https://devfeed.tech/tags/frontend-real-time.md>), [jamstack](<https://devfeed.tech/tags/jamstack.md>), [llms](<https://devfeed.tech/tags/llms.md>), [perplexity](<https://devfeed.tech/tags/perplexity.md>), [sdks-extensions-plugins](<https://devfeed.tech/tags/sdks-extensions-plugins.md>), [seo](<https://devfeed.tech/tags/seo.md>)

### AI overview

The article's six-month server-log findings support two priorities: clean, server-rendered HTML for chat assistants, and Markdown content negotiation for coding agents that request it. In the authors' logs, chat fetchers mostly read HTML; over half of agent reads in September and October were Markdown, nearly all from Claude Code. Serve `.md` routes and negotiate `Accept: text/markdown` with `Vary: Accept`, correct q-value handling, and `406` responses for unsupported formats. Keep HTML semantically structured and lightweight, and measure requests in server logs because AI crawlers often do not run JavaScript. The authors found little AI use of `llms.txt` (ChatGPT fetched it once on their site), so they recommend it only as a low-cost, low-expectation addition. JSON-LD may help search indexes and shopping systems indirectly, though tests cited found most models could not read it directly. Check crawler access and distinguish provider crawlers from on-demand fetchers; results vary by audience and client.

### Source excerpt

Most AI SEO advice is unproven. We tested what ChatGPT, Claude, and Perplexity actually read on our own site. Six techniques that make a site agent-friendly and visible to LLMs, eight that don't, and the metrics to tell the difference.

## How We Optimized Browser Use for Deep Agents

DevFeed: [How We Optimized Browser Use for Deep Agents](<https://devfeed.tech/articles/how-we-optimized-browser-use-for-deep-agents-116377.md>)

Original publisher: [Read original article](<https://www.langchain.com/blog/how-we-optimized-browser-use-for-deep-agents>)

Author: LangChain Accounts

Published: 2026-11-03T22:30:50Z

Content type: article

Language: en

Sources: [LangChain Blog](<https://devfeed.tech/sources/langchain-blog.md>)

Topics: [Agent Browser](<https://devfeed.tech/topics/agent-browser.md>), [agentic browsing](<https://devfeed.tech/topics/agentic-browsing.md>), [OpenTelemetry](<https://devfeed.tech/topics/opentelemetry.md>), [browser](<https://devfeed.tech/topics/browser.md>), [Selenium](<https://devfeed.tech/topics/selenium.md>)

Tags: [agent](<https://devfeed.tech/tags/agent.md>), [agentic](<https://devfeed.tech/tags/agentic.md>), [agents](<https://devfeed.tech/tags/agents.md>), [api](<https://devfeed.tech/tags/api.md>), [apis](<https://devfeed.tech/tags/apis.md>), [benchmark](<https://devfeed.tech/tags/benchmark.md>), [benchmarks](<https://devfeed.tech/tags/benchmarks.md>), [browser](<https://devfeed.tech/tags/browser.md>), [browser-agents](<https://devfeed.tech/tags/browser-agents.md>), [browser-tools](<https://devfeed.tech/tags/browser-tools.md>), [browserbase](<https://devfeed.tech/tags/browserbase.md>), [browsers](<https://devfeed.tech/tags/browsers.md>), [browsing](<https://devfeed.tech/tags/browsing.md>), [cdp](<https://devfeed.tech/tags/cdp.md>), [chrome](<https://devfeed.tech/tags/chrome.md>), [chrome-devtools](<https://devfeed.tech/tags/chrome-devtools.md>), [claude](<https://devfeed.tech/tags/claude.md>), [evals](<https://devfeed.tech/tags/evals.md>), [langsmith](<https://devfeed.tech/tags/langsmith.md>), [traces](<https://devfeed.tech/tags/traces.md>)

### AI overview

The team selected Browserbase's Stagehand after comparing seven browser toolsets, then used LangSmith traces and evals to refine it. In their benchmarks, Stagehand passed 90% of WebVoyager tasks for $1.98 total, while Playwright MCP passed 78% for $26.71 and took three times as long per task. Stagehand's code-based `run` batches browser actions, keeping intermediate page data out of the model context; its `snapshot` returns a compact accessibility tree with IDs for interactive elements; and running actions in the browser reduces round trips. A separate harness change made fixed waits end once the page had been quiet for 500 ms. On HardBench, combined with structured answers and a model-call timeout, this reduced median time from 182 to 124 seconds while accuracy stayed within noise. The reported results cover the stated benchmarks and setups; the suggested optimization process is to inspect traces, change tools or harness behavior, and check changes with evals.

### Source excerpt

We compared seven browser toolsets for Deep Agents, picked Stagehand from Browserbase, and tuned it with LangSmith traces and evals. Here's what we learned.

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

DevFeed: [When 200 OK Is Not an Answer: Silent Fetch Failures in Agentic Web Content Acquisition](<https://devfeed.tech/articles/when-200-ok-is-not-an-answer-silent-fetch-failures-in-agentic-web-content-acquisition-113856.md>)

Original publisher: [Read original article](<https://arxiv.org/abs/2611.00884>)

Author: Shuyang Zhang (The Hong Kong Polytechnic University)

Published: 2026-11-03T05:00:00Z

Content type: news

Language: en

Sources: [arXiv: Networking](<https://devfeed.tech/sources/arxiv-networking.md>)

Topics: [agentic browsing](<https://devfeed.tech/topics/agentic-browsing.md>), [AI Bots](<https://devfeed.tech/topics/ai-bots.md>)

Tags: [agentic](<https://devfeed.tech/tags/agentic.md>), [agents](<https://devfeed.tech/tags/agents.md>), [cs-cl](<https://devfeed.tech/tags/cs-cl.md>), [cs-ir](<https://devfeed.tech/tags/cs-ir.md>), [cs-ni](<https://devfeed.tech/tags/cs-ni.md>), [fetch](<https://devfeed.tech/tags/fetch.md>), [web](<https://devfeed.tech/tags/web.md>)

### AI overview

In a study of 10,307 URLs across 1,507 sites, 20.9% of pages returned to a static client with HTTP 200 lacked the page's main content, including app shells, bot challenges, block pages, errors, placeholders, and partial content. For ExpertQA citations, treating every 200 as usable content overstated live-checkable claim-source pairs by 12.2 percentage points and raised apparent silent citation drift from 5.8% to 21.9%. A held-out combination of text length, comparison with random-path probes, and redirects to the site root detected 86% of reviewed unusable responses at about a 5% false-alarm rate. The best content-reading detector, a text classifier, improved recall by five to seven points but had more than twice the false-alarm rate. The authors recommend reporting fetch outcomes explicitly; the evidence here is the supplied abstract and excerpt.

### Source excerpt

arXiv:2611.00884v1 Announce Type: cross Abstract: 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.

## ANGLER: Verifier-Grounded Benchmarking with an Adaptive Attacker for Web Agents

DevFeed: [ANGLER: Verifier-Grounded Benchmarking with an Adaptive Attacker for Web Agents](<https://devfeed.tech/articles/angler-verifier-grounded-benchmarking-with-an-adaptive-attacker-for-web-agents-111217.md>)

Original publisher: [Read original article](<https://arxiv.org/abs/2610.41272>)

Author: Yini Huang, Aiwen LU, Zhen Sun, Yi Han, Yuzhi Zhao, Jiaheng Wei

Published: 2026-11-02T05:00:00Z

Content type: article

Language: en

Sources: [arXiv: Artificial Intelligence](<https://devfeed.tech/sources/arxiv-artificial-intelligence.md>), [arXiv: Cryptography and Security](<https://devfeed.tech/sources/arxiv-cryptography-and-security.md>)

Topics: [screensuite](<https://devfeed.tech/topics/screensuite.md>), [agentic browsing](<https://devfeed.tech/topics/agentic-browsing.md>), [ai-assisted attacks](<https://devfeed.tech/topics/ai-assisted-attacks.md>), [AI Bots](<https://devfeed.tech/topics/ai-bots.md>)

Tags: [adaptive](<https://devfeed.tech/tags/adaptive.md>), [agents](<https://devfeed.tech/tags/agents.md>), [attack](<https://devfeed.tech/tags/attack.md>), [attacks](<https://devfeed.tech/tags/attacks.md>), [benchmark](<https://devfeed.tech/tags/benchmark.md>), [benchmarking](<https://devfeed.tech/tags/benchmarking.md>), [cs-ai](<https://devfeed.tech/tags/cs-ai.md>), [cs-cr](<https://devfeed.tech/tags/cs-cr.md>), [red-teaming](<https://devfeed.tech/tags/red-teaming.md>), [reef](<https://devfeed.tech/tags/reef.md>), [security](<https://devfeed.tech/tags/security.md>), [security-vulnerabilities](<https://devfeed.tech/tags/security-vulnerabilities.md>), [verification](<https://devfeed.tech/tags/verification.md>), [web](<https://devfeed.tech/tags/web.md>)

### AI overview

REEF evaluates web-agent security through state-based verification across 9 websites and 5 attack types, pairing benign tasks with malicious objectives delivered at user or environment level. Its verifiers feed execution outcomes into ANGLER, which adaptively mutates and selects LLM-generated lures, prioritizing near-miss attacks instead of using a fixed set. Across 8 VLM-based agents, ANGLER achieved a 20.36% attack success rate versus 15.24% without adaptation. On the same benign tasks, generated attacks lowered benign task success from 61.9% in attack-free runs to 44.4%.

### Source excerpt

arXiv:2610.41272v1 Announce Type: new Abstract: Web agents are increasingly deployed to complete tasks autonomously on real websites, where adversarially injected content can induce them to perform unauthorized actions. Existing evaluations primarily use fixed attacks or focus on a single attack type, and evaluate attacks at the text level rather than through real state changes, thus underestimating how vulnerable agents are to diverse attacks. To address the gap, we introduce REEF, a benchmark of red-teaming executable environments with faithful verification for measuring the security vulnerabilities of web agents. REEF covers 9 verifier-grounded websites spanning travel, commerce, productivity, media, and communication, and pairs every authentic benign task with a malicious objective delivered through 5 attack types at both the user and the environment level. REEF's state-based verifiers provide the execution feedback that drives ANGLER, an adaptive near-miss-guided lure evolution framework for red-teaming. Rather than evaluating a fixed attack set, ANGLER uses outcomes from prior rollouts to guide LLM-based attack mutation and candidate selection, prioritizing near-miss seeds during refinement. Across 8 VLM-based web agents, ANGLER achieves an end-to-end attack success rate (ASR) of 20.36%, compared with 15.24% for the non-adaptive setting. Compared with attack-free execution on the same benign tasks, attacks generated by ANGLER reduce benign task success rate (BSR) from 61.9% to 44.4%.

## AgentShield: Fortifying Web Applications Against AI Agents and Agentic Browsers

DevFeed: [AgentShield: Fortifying Web Applications Against AI Agents and Agentic Browsers](<https://devfeed.tech/articles/agentshield-fortifying-web-applications-against-ai-agents-and-agentic-browsers-111215.md>)

Original publisher: [Read original article](<https://arxiv.org/abs/2610.41181>)

Author: Moiz Ali, Nafis Karim, Rownak Islam, Jyotirmay Chauhan, Dhruv Kuchhal, Jason Polakis

Published: 2026-11-02T05:00:00Z

Content type: article

Language: en

Sources: [arXiv: Cryptography and Security](<https://devfeed.tech/sources/arxiv-cryptography-and-security.md>)

Topics: [agentic browsing](<https://devfeed.tech/topics/agentic-browsing.md>), [AI Bots](<https://devfeed.tech/topics/ai-bots.md>), [Agentic AI](<https://devfeed.tech/topics/what-is-agentic-ai.md>), [frontend security](<https://devfeed.tech/topics/frontend-security.md>)

Tags: [agentic](<https://devfeed.tech/tags/agentic.md>), [agents](<https://devfeed.tech/tags/agents.md>), [ai](<https://devfeed.tech/tags/ai.md>), [ai-agents](<https://devfeed.tech/tags/ai-agents.md>), [bot-defenses](<https://devfeed.tech/tags/bot-defenses.md>), [browser-agents](<https://devfeed.tech/tags/browser-agents.md>), [browsers](<https://devfeed.tech/tags/browsers.md>), [cs-cr](<https://devfeed.tech/tags/cs-cr.md>), [detection](<https://devfeed.tech/tags/detection.md>), [evaluation](<https://devfeed.tech/tags/evaluation.md>), [prompt-injection](<https://devfeed.tech/tags/prompt-injection.md>), [web](<https://devfeed.tech/tags/web.md>), [web-applications](<https://devfeed.tech/tags/web-applications.md>)

### AI overview

The abstract reports that many anti-bot defenses built for traditional automation struggle to detect AI browser agents. AgentShield aims to distinguish agents from people by using dynamic forms that reveal content progressively and collect detailed interaction telemetry tied to the agents' perception-reasoning-action loop. The authors also describe Extinction-induced Policy Drift: repeated obstruction can lead agents to relax safeguards and behave unsafely. They propose defensive prompt injections to mitigate agent-driven content extraction. The supplied text does not include detailed measurements or deployment specifics.

### Source excerpt

arXiv:2610.41181v1 Announce Type: new Abstract: LLM-powered browser agents are transforming web automation by autonomously interacting with websites and executing complex workflows. However, the effectiveness of existing anti-bot defenses against these systems remains unclear. We present a systematic empirical evaluation of modern agentic browsers against widely deployed anti-bot mechanisms, characterizing their capabilities, limitations, and behavioral differences from humans. Our results show that many defenses designed for traditional automation struggle to reliably detect autonomous AI-driven agents. Accordingly, we introduce AgentShield, a behavioral fingerprinting framework that targets the Perception-Reasoning-Action (PRA) loop underlying agentic user clients. By leveraging dynamic web forms that progressively reveal content and capture fine-grained interaction telemetry, AgentShield exposes behavioral artifacts that enable differentiation between human users and autonomous agents. We further demonstrate Extinction-induced Policy Drift (EPD), a novel attack in which repeated obstruction of task completion causes agents to relax their safeguards and engage in unsafe behaviors. Finally, we present defensive prompt injection techniques that exploit agent-specific weaknesses to mitigate emerging forms of autonomous web abuse that rely on content extraction. Our results highlight both the shortcomings of current web defenses against agentic user clients and the feasibility of practical detection mechanisms tailored to this new generation of web automation.

## DynaRev: Benchmarking Web Agents under Dynamic Goal Revision

DevFeed: [DynaRev: Benchmarking Web Agents under Dynamic Goal Revision](<https://devfeed.tech/articles/dynarev-benchmarking-web-agents-under-dynamic-goal-revision-108268.md>)

Original publisher: [Read original article](<https://arxiv.org/abs/2610.40133>)

Author: Shiqi Zhang, Guochen Zhou, Zesheng Wei, Hai Zhao, Zhisong Zhang

Published: 2026-10-30T04:00:00Z

Content type: article

Language: en

Sources: [arXiv: Artificial Intelligence](<https://devfeed.tech/sources/arxiv-artificial-intelligence.md>)

Topics: [AI Agent](<https://devfeed.tech/topics/ai-agent.md>), [agentic browsing](<https://devfeed.tech/topics/agentic-browsing.md>), [test](<https://devfeed.tech/topics/test.md>), [Simulation](<https://devfeed.tech/topics/simulation.md>)

Tags: [agents](<https://devfeed.tech/tags/agents.md>), [benchmark](<https://devfeed.tech/tags/benchmark.md>), [benchmarking](<https://devfeed.tech/tags/benchmarking.md>), [browser](<https://devfeed.tech/tags/browser.md>), [cs-ai](<https://devfeed.tech/tags/cs-ai.md>), [data](<https://devfeed.tech/tags/data.md>), [efficiency](<https://devfeed.tech/tags/efficiency.md>), [evaluation](<https://devfeed.tech/tags/evaluation.md>), [execution](<https://devfeed.tech/tags/execution.md>), [goals](<https://devfeed.tech/tags/goals.md>), [handling](<https://devfeed.tech/tags/handling.md>), [interaction](<https://devfeed.tech/tags/interaction.md>), [milestone](<https://devfeed.tech/tags/milestone.md>), [milestones](<https://devfeed.tech/tags/milestones.md>), [model](<https://devfeed.tech/tags/model.md>), [models](<https://devfeed.tech/tags/models.md>), [performance](<https://devfeed.tech/tags/performance.md>), [requirements](<https://devfeed.tech/tags/requirements.md>), [revision](<https://devfeed.tech/tags/revision.md>), [revisions](<https://devfeed.tech/tags/revisions.md>), [web](<https://devfeed.tech/tags/web.md>)

### AI overview

DynaRev evaluates how web agents handle changing user goals through 181 tasks that deliver revisions at semantic milestones. It covers additive changes, unconditional overrides, and runtime-conditioned overrides, and scores correctness, responsiveness, and efficiency using agent trajectories. Evaluations of five models found that agents often met only some requirements, with results differing by task source, revision type, and model. Analyses report mixed effects of delivering revisions dynamically: rewriting instructions while preserving browser state improved outcomes, while resetting the browser hurt them. The supplied abstract describes the benchmark and findings but does not provide detailed results.

### Source excerpt

arXiv:2610.40133v1 Announce Type: new Abstract: 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.

## WAAA! Web Adversaries Against Web Agents

DevFeed: [WAAA! Web Adversaries Against Web Agents](<https://devfeed.tech/articles/waaa-web-adversaries-against-web-agents-108991.md>)

Original publisher: [Read original article](<https://arxiv.org/abs/2605.05509>)

Author: Sohom Datta, Alex Nahapetyan, William Enck, Alexandros Kapravelos

Published: 2026-10-30T04:00:00Z

Content type: article

Language: en

Sources: [arXiv: Cryptography and Security](<https://devfeed.tech/sources/arxiv-cryptography-and-security.md>)

Topics: [LLM security](<https://devfeed.tech/topics/llm-security.md>), [agentic browsing](<https://devfeed.tech/topics/agentic-browsing.md>), [Agent Browser](<https://devfeed.tech/topics/agent-browser.md>), [AI Bots](<https://devfeed.tech/topics/ai-bots.md>), [prompt injection](<https://devfeed.tech/topics/prompt-injection.md>), [Large Language Model](<https://devfeed.tech/topics/llm.md>)

Tags: [actions](<https://devfeed.tech/tags/actions.md>), [agentic](<https://devfeed.tech/tags/agentic.md>), [agentic-browsing](<https://devfeed.tech/tags/agentic-browsing.md>), [agents](<https://devfeed.tech/tags/agents.md>), [attacks](<https://devfeed.tech/tags/attacks.md>), [browser](<https://devfeed.tech/tags/browser.md>), [browsers](<https://devfeed.tech/tags/browsers.md>), [components](<https://devfeed.tech/tags/components.md>), [cs-cr](<https://devfeed.tech/tags/cs-cr.md>), [failure](<https://devfeed.tech/tags/failure.md>), [indirect-prompt-injection](<https://devfeed.tech/tags/indirect-prompt-injection.md>), [large-language-models-llms](<https://devfeed.tech/tags/large-language-models-llms.md>), [llms](<https://devfeed.tech/tags/llms.md>), [security](<https://devfeed.tech/tags/security.md>), [social-engineering](<https://devfeed.tech/tags/social-engineering.md>), [social-engineering-attacks](<https://devfeed.tech/tags/social-engineering-attacks.md>), [taxonomy](<https://devfeed.tech/tags/taxonomy.md>), [threats](<https://devfeed.tech/tags/threats.md>), [web](<https://devfeed.tech/tags/web.md>)

### AI overview

The paper argues that agentic browser security models miss traditional web social engineering, leaving agents vulnerable when they cannot distinguish user task steps from malicious page content. It extends the See->Act model to cover browser components, defines 20 attacks across web and LLM threats, and implements 18. The authors report that 10 traditional web threats can reappear in amplified forms and that 14 attacks reproduced across four major LLM models from multiple vendors. They identify five failure modes and argue that agentic browsers need architectural changes before they are ready for the current web.

### Source excerpt

arXiv:2605.05509v2 Announce Type: replace Abstract: Large language models (LLMs) are increasingly being integrated into web browsers to create agentic browsing systems that execute actions on behalf of the user. Prior work considering the security of agentic browsers focuses on indirect prompt-injection attacks. However, by failing to consider traditional web attacks, previous agentic browser threat models have a blind spot to web social engineering attacks originally designed to trick humans. In this paper, we propose the first web-focused threat model for agentic browsers and use it to derive a taxonomy of 20 attacks across both the web and LLM space, and implement 18 of the attacks. Our threat model extends the original See$\rightarrow$Act browser agent model to account for all components of a browser, and frames the agent as a confused deputy unable to distinguish task steps from traditional web attacks. We show that 10 web threats can reemerge often in amplified forms once an agent can be influenced by untrusted page content. We further conduct a generalizability study on 14 of the 20 attacks, showing that our attacks reproduce across 4 major LLM models spanning multiple vendors. We show that agentic browsers exhibit five major failure modes when facing traditional and LLM web threats, demonstrating the need to rearchitect agentic browsers before they are ready for the current web.

## WebAnything: From the Wild Web to Scalable Agent Environments

DevFeed: [WebAnything: From the Wild Web to Scalable Agent Environments](<https://devfeed.tech/articles/webanything-from-the-wild-web-to-scalable-agent-environments-107965.md>)

Original publisher: [Read original article](<https://arxiv.org/abs/2610.37786>)

Author: Tianci Xie, Zikun Cai, Chaoyun Zhang, Pinxue Guo, Bingxuan Li, Rui Zheng, Hang Yan

Published: 2026-10-30T04:00:00Z

Content type: news

Language: en

Sources: [arXiv: Artificial Intelligence](<https://devfeed.tech/sources/arxiv-artificial-intelligence.md>)

Topics: [agentic browsing](<https://devfeed.tech/topics/agentic-browsing.md>), [AI Chat](<https://devfeed.tech/topics/ai-chat.md>), [Web](<https://devfeed.tech/topics/web.md>)

Tags: [abstract](<https://devfeed.tech/tags/abstract.md>), [agent](<https://devfeed.tech/tags/agent.md>), [agents](<https://devfeed.tech/tags/agents.md>), [artifacts](<https://devfeed.tech/tags/artifacts.md>), [automated](<https://devfeed.tech/tags/automated.md>), [browser](<https://devfeed.tech/tags/browser.md>), [builder](<https://devfeed.tech/tags/builder.md>), [cs-ai](<https://devfeed.tech/tags/cs-ai.md>), [domains](<https://devfeed.tech/tags/domains.md>), [environments](<https://devfeed.tech/tags/environments.md>), [evaluation](<https://devfeed.tech/tags/evaluation.md>), [fine-tuning](<https://devfeed.tech/tags/fine-tuning.md>), [interaction](<https://devfeed.tech/tags/interaction.md>), [qwen3-5](<https://devfeed.tech/tags/qwen3-5.md>), [site](<https://devfeed.tech/tags/site.md>), [source](<https://devfeed.tech/tags/source.md>), [tasks](<https://devfeed.tech/tags/tasks.md>), [training](<https://devfeed.tech/tags/training.md>), [web](<https://devfeed.tech/tags/web.md>)

### AI overview

WebAnything reconstructs reusable, resettable website environments from real URLs, aiming to support reproducible web-agent training and evaluation without per-site manual implementation after initial setup and authorization. Its Explorer-Builder gathers browser-visible structures, resources, and interaction outcomes; an independent verifier checks frozen candidates and directs local repairs or targeted recaptures. The authors report 244 artifacts across 14 application domains and 53 categories, supporting 12,984 executed and accepted tasks. In a blind audit against human-assisted WebHarbor, WebAnything scored higher in aggregate on three fidelity metrics, with 12 wins, two ties, and no losses. Fine-tuning Qwen3.5-9B on 5,856 interaction trajectories improved reported success or average scores on REAL, Odysseys-Easy, and WebArena-Verified under a fixed harness. The supplied abstract does not detail the fidelity metrics or evaluation methods.

### Source excerpt

arXiv:2610.37786v1 Announce Type: new Abstract: 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

[Next page](<https://devfeed.tech/topics/agentic-browsing.md?cursor=WyIyMDI2LTEwLTMwVDA0OjAwOjAwKzAwOjAwIiwgIjk3ZWNmMDA2LWQ1YWYtNGJiNC1hNzBmLTQyMGZlYWNhNzMxMCJd>)