Pith. sign in

REVIEW 4 major objections 4 minor 89 references

Controlled experiments with ten popular AI assistants show that six retrieved pages their own robots.txt rules disallowed, often without ever fetching the robots.txt file, so the standard cannot be assumed to govern inference-time web acces

Reviewed by Pith at T0; open to challenge. T0 means a machine referee read the full paper against a public rubric. the ladder, T0–T4 →

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗

T0 review · deepseek-v4-flash

2026-08-02 02:02 UTC pith:TTXYS4U5

load-bearing objection A genuinely new controlled measurement of AI-assistant robots.txt compliance, with a real but acknowledged prompt confound; worth a serious referee and conditional acceptance. the 4 major comments →

arxiv 2607.14447 v2 pith:TTXYS4U5 submitted 2026-07-16 cs.CY

Do Generative AI Assistants Respect robots.txt? Tracing Web Access Beyond Visible Answers

Gabriel Lopez-Fonseca , David Rodriguez , Stefan Bechtold , Jose M. Del Alamo This is my paper
classification cs.CY
keywords robots.txtAI assistantsweb crawlinginference-time retrievaluser-agent attributioncompliance measurementretrieval-augmented generationweb governance
verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

This paper asks whether AI assistants that fetch web pages while answering a user question actually respect the access rules website owners publish in robots.txt. In 200 controlled trials across ten assistants, the authors set up their own server, made pages that were allowed or disallowed for everyone or only for a specific assistant's user-agent, and used hidden codes to prove whether a page was really fetched. They found no uniform behavior: Claude and Mistral followed the allow/disallow pattern, while Copilot, DeepSeek, Diffy Chat, Gemini, Grok, and Qwen retrieved disallowed pages, and most of those never requested robots.txt at all. Several assistants also identified themselves with generic browser-like user-agents, making it hard for a site to block them selectively. The authors conclude that robots.txt cannot be relied on to govern this form of access, and that website owners may need stronger enforcement tools.

Core claim

The paper's central finding is that inference-time web retrieval by AI assistants is not reliably governed by the Robots Exclusion Protocol. Using a controlled domain and server logs, the authors observed that under an explicit Disallow rule, six of ten assistants still fetched the forbidden page (Copilot, DeepSeek, Diffy Chat, Gemini, Grok, Qwen), and four of those (DeepSeek, Gemini, Grok, Qwen) returned the secret code embedded in the page, showing the content reached the model. None of these six requested robots.txt during the active trials. In contrast, Claude consistently fetched only allowed pages and told the user when robots.txt blocked access; Mistral similarly avoided disallowed pa

What carries the argument

The experimental apparatus is a two-phase controlled measurement. Phase 1 identifies, for each assistant, a configuration that actually produces a server-visible HTTP request to a target page and records the user-agent string the assistant exposes. Phase 2 uses a fresh domain whose robots.txt implements four conditions (allow all, disallow all, allow assistant-specific user-agent, disallow assistant-specific user-agent). Each target page carries a unique HMAC-derived secret code; server logs record robots.txt requests, page requests, and user-agents, and the assistant's answer is checked for the code. This lets the authors separate whether a page was fetched from whether the answer used it,

Load-bearing premise

The load-bearing assumption is that the browser-identification string (user-agent) seen in the first phase belongs to the same retrieval component that fetches pages in the second phase, so the assistant product, not a separate crawler, is responsible for the disallowed accesses.

What would settle it

A single controlled re-run in which the server records the full request path for one assistant: if the page that returns the secret code from a disallowed folder is requested by an IP or user-agent different from the one seen in Phase 1 (e.g., a known search-index crawler rather than the assistant's live browser tool), the paper's attribution of non-compliance to that assistant would collapse. Concretely, serving the disallowed page only when the request carries the Phase-1 user-agent, and returning an error to all other user-agents, would show whether the assistant's own component is the one

Watch this falsifier. Get emailed when new claim-graph text bears on it.

If this is right

  • Website owners cannot assume robots.txt alone protects their content from assistant-mediated retrieval; they may need rate limiting, bot verification, or active blocking.
  • Assistant-specific robots.txt rules are weakened whenever an assistant uses a generic user-agent, because the site cannot attribute the request to a particular provider.
  • Compliance audits must include positive controls: a system that fetches nothing in allowed conditions is not 'compliant' with disallow rules, it is simply failing retrieval.
  • Providers' stated commitments to honor rights reservations are not automatically fulfilled by their production assistant components; the observed variation across products shows the gap.
  • Server-side access and user-visible answers are different signals; judging assistants by answers alone would misclassify Copilot (fetches but doesn't answer) and ChatGPT (answers nothing).

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • If this pattern generalizes, the practical endpoint is an arms race: as generic user-agents defeat passive controls, site owners shift to CAPTCHAs, IP blocks, and other active measures, increasing friction for human visitors and fragmenting the open web.
  • A direct follow-up test would vary the prompt: the paper notes its ownership-stating prompt may have increased retrieval willingness; repeating the trials with a neutral prompt could quantify how much of the observed non-compliance is prompt-induced.
  • The secret-code method could be extended to measure downstream use: embed unique markers, then later query the same assistant for facts only present on disallowed pages, testing whether retrieved content entered training or long-term memory.
  • The distinction between retrieval and visible answers suggests that future audits should instrument the server side rather than evaluating assistant outputs, and that user-facing citation checks are not a reliable compliance signal.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit.

Referee Report

4 major / 4 minor

Summary. The paper reports a controlled empirical study of ten widely used AI assistants with advertised web-search/browsing capabilities. In Phase 1 the authors identify, for each assistant, a configuration that produces observable web retrieval and record the user-agent exposed. In Phase 2 they test each assistant under four robots.txt conditions — allow all, disallow all, allow the assistant-specific user-agent, and disallow that user-agent — using server-side logs and HMAC-secret-coded pages to distinguish actual retrieval from answer correctness, for a total of 200 trials. The main findings are that some assistants (Claude, Mistral, largely Perplexity) follow the expected allow/disallow pattern, whereas others (Copilot, DeepSeek, Diffy Chat, Gemini, Grok, Qwen) access pages under disallow conditions without requesting robots.txt, often under generic user-agents. The paper also documents cases where retrieval and user-visible answers diverge, and it draws implications for web governance, the EU DSM Directive, the EU AI Act, and downstream training use.

Significance. If the results hold, the paper is a valuable and timely empirical contribution. The server-side HMAC/log design is a solid way to verify actual page access, and the four-condition design with both allow and disallow conditions correctly avoids mistaking retrieval failure for compliance. The study is observant rather than model-derived, so it has no circular-reasoning burden. The paper also provides useful positive controls and a reproducible methodology, and it makes its data availability plan explicit. The main claims — that inference-time web access by AI assistants is not reliably governed by robots.txt and that server-side access and answer correctness can diverge — are falsifiable and important for the AI-governance and web-measurement communities.

major comments (4)
  1. [Section 4.3 / Section 8] The retrieval prompt used in all trials is 'Get the contents from [URL]. I manage the server and will know if you are really accessing it'. The ownership assertion is a potential authorization cue: a helpful assistant may treat an explicit owner request as overriding robots.txt. The Limitations section acknowledges this possibility but does not test a neutral prompt. Because the headline conclusion ('robots.txt cannot be assumed to reliably govern inference-time web access by AI assistants', Section 1) is about general system behavior, this confound is load-bearing. I recommend adding a neutral-prompt condition (e.g., 'Please fetch the content at [URL]') across the four robots.txt conditions, at least for the assistants that accessed disallowed pages (DeepSeek, Gemini, Grok, Qwen). Without that, the conclusions should be scoped to behavior under this specific ownership-asserting prompt.
  2. [Section 2.6 / Section 6.1 / Table 4] The study assumes that the user-agent observed in Phase 1 identifies the same retrieval subsystem that operates in Phase 2. The paper itself stresses that assistant retrieval is layered and that different components may use different user-agents. For assistants with generic user-agents, the only link between a log entry and the assistant is the page-version range embedded in the URL. This does not rule out an indexing crawler, a third-party fetcher, or another internal component accessing the page outside the live-browsing path. The assistant-specific compliance claims in Tables 1 and 4 rest on this untested assumption. Please report per-trial request timestamps relative to the prompt and, where possible, use per-trial unique tokens in the URL to verify that the observed access is the live retrieval triggered by the prompt.
  3. [Section 6.2.2 / Appendix A.11] Grok's 48–52 target-page accesses per five-trial condition are attributed to the assistant, but the log excerpts show requests from many different IPs, multiple browser user-agents, and even a 404 for a trailing-dot URL. This pattern is more consistent with a distributed fetching service or multiple internal components than with a single live-browsing agent. The paper should either verify that all these requests are under the assistant's control (e.g., via timing correlation with the prompt and IP-ownership checks) or present the amplification as an artifact of the retrieval pipeline rather than as deliberate repeated access by the assistant product.
  4. [Section 6.2.2 / Section 6.2.4] The claim that DeepSeek, Gemini, Grok, and Qwen accessed restricted pages 'without requesting robots.txt' rests on the absence of observed robots.txt requests during active trials and in historical logs. For Gemini, however, the paper reports unverified Google-user-agent requests that were not attributed. The absence of a robots.txt request in the logs could also result from a different user-agent or a cached fetch outside the checked window. Please specify the exact time window around each trial in which robots.txt absence was checked, and describe the IP/UA filtering used to associate a robots.txt request with the assistant being tested.
minor comments (4)
  1. [Throughout] There are numerous typos: 'robotx.txt' appears in Section 2, Section 4.3, Section 6.2 (table header), and Section 8; the title in the provided text appears as 'DoGenerative AI Assistants' without spacing; and 'robotx.txtexperiment' appears in Section 5.2. Please correct these.
  2. [Section 5.2 / Table 1] Table 1 lists Copilot's user-agent as 'Diffbot-User', but Section 6.1 classifies Copilot as having a generic/non-identifiable user-agent. Please clarify why a Diffbot-User string is considered non-attributable to the assistant product.
  3. [Figure 5] The passive-monitoring time series would be easier to interpret if the figure clearly labeled which line/color corresponds to which assistant and included a legend; currently the narrative in Section 6.2.4 is needed to understand the plot.
  4. [Section 6.2 / Table 4] With five trials per condition, exact binomial confidence intervals would help readers judge whether zero/one access counts are distinguishable from chance-level retrieval failures. The paper's main patterns are probably robust, but adding intervals (or at least exact counts per trial in the supplement) would strengthen the presentation.

Circularity Check

0 steps flagged

No significant circularity: the study is observational, with independent server-side measurements and no fitted or self-citational derivation chain.

full rationale

This paper is an empirical measurement study, not a derivation. Its central claim—that some AI assistants access robots.txt-disallowed pages at inference time—rests on server-side logs and page-specific HMAC secret codes, which are independent of the paper's own assumptions. There are no fitted parameters or equations whose outputs are equivalent to their inputs by construction. The Phase 1 user-agents are used to build Phase 2 robots.txt groups, but the Phase 2 dependent variable is actual access behavior under those rules; the user-agent is a measured input, not a predicted output, so the access results are not forced by the experimental design. The paper contains no self-citations by its authors: the references are external work on benchmarks, crawler compliance, and legal frameworks. No uniqueness theorem, ansatz, or prior result by the same authors is invoked to justify the conclusions. The Limitations section candidly notes that the prompt wording ('I manage the server...') 'may have increased some assistants' willingness to retrieve the page' and that the black-box systems prevent identification of which internal component performed the retrieval; these are validity threats that limit generalization, but they are not circular-reasoning steps. The paper explicitly frames its findings as behavior under this standardized prompt, which is an appropriate scoping rather than a circularity. Accordingly, the circularity score is 0.

Axiom & Free-Parameter Ledger

1 free parameters · 4 axioms · 0 invented entities

No free parameters are fitted to data; the only hand-chosen numeric is the trial count. The study relies on standard web-protocol and logging assumptions plus the load-bearing user-agent attribution assumption. No new entities are postulated.

free parameters (1)
  • Trials per condition = 5
    Hand-chosen sample size per condition without a power analysis; adequate for detecting binary compliance patterns but limits precision for mixed results (e.g., Perplexity accessing 1/5 in Disallow All).
axioms (4)
  • domain assumption RFC 9309 is the applicable standard for robots.txt interpretation
    The study defines compliance as consistent behavior across the four conditions per RFC 9309; the legal and technical framing depends on this standard being the relevant one for the tested systems.
  • domain assumption Server-side logs and HMAC secret codes validly prove page access
    The paper infers retrieval from HTTP log entries plus correct secret codes in answers; a black-box assistant could in principle obtain content through an unobserved path, though the controlled domain makes this unlikely.
  • ad hoc to paper The user-agent observed in Phase 1 identifies the same retrieval subsystem in Phase 2
    Section 2.6 acknowledges the layered architecture (search-index crawler vs. live browser tool vs. assistant-side fetcher); the attribution of non-compliance to a specific assistant assumes user-agent stability across phases and components, which is load-bearing.
  • domain assumption Assistant behavior did not change across the experimental window
    Section 8 acknowledges that model updates, infrastructure changes, and safety filters could change behavior without notice; the results are temporally bounded.

pith-pipeline@v1.3.0-alltime-deepseek · 30869 in / 12064 out tokens · 113897 ms · 2026-08-02T02:02:50.446754+00:00 · methodology

0 comments
read the original abstract

AI assistants increasingly retrieve web content at inference time to provide fresh and grounded answers, yet it remains unclear whether these search-augmented capabilities respect website-owner restrictions expressed through robots$.$txt. We present a controlled empirical study of ten widely used AI assistants with advertised web-search capabilities. For each assistant, we first identify a configuration that actually produces observable web-browsing behavior and record the user-agent exposed during retrieval. We then evaluate compliance with controlled robots$.$txt rules across four complementary conditions: allowed for all user-agents, disallowed for all user-agents, allowed only for the assistant-specific user-agent, and disallowed only for that user-agent. Using server-side logs and secret codes embedded in target pages, we distinguish actual page access from user-visible answer correctness across 200 trials. Our results show substantial variation across assistants. Some systems followed the expected allowed/disallowed access pattern, whereas others accessed restricted resources without requesting robots$.$txt or used generic user-agents that complicated attribution. We also find that retrieval behavior and answer correctness can diverge: assistants may access pages without surfacing the retrieved content, or fail to access even allowed resources. These findings raise broader legal and governance concerns about whether AI-assisted web access adequately respects content owners' rights and restrictions. Furthermore, our observations provide valuable insight into the growing erosion of traditional web governance protocols, highlighting the urgent need for updated, enforceable standards that guarantee publisher autonomy in the age of search-augmented AI assistants.

Figures

Figures reproduced from arXiv: 2607.14447 by David Rodriguez, Gabriel Lopez-Fonseca, Jose M. del Alamo, Stefan Bechtold.

Figure 2
Figure 2. Figure 2: 4.3 AI Assistant Selection and Prompt Design Assistant selection. The selection of AI assistants was based on the top-ranked systems in the LMArena Search Arena Leaderboard, which ranks LLMs with integrated web-search capabilities (LMArena, 2026). The leaderboard was accessed on January 29, 2026. The selected leading assistants were Gemini, ChatGPT, Grok, Claude, Perplexity, and Diffy Chat. We additionally… view at source ↗
Figure 1
Figure 1. Figure 1: Experimental interaction between the user, the AI assistant, and one of our controlled websites. The figure [PITH_FULL_IMAGE:figures/full_fig_p008_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Experimental setting of our Controlled Web Server. The figure illustrates the steps of the flow of how a user [PITH_FULL_IMAGE:figures/full_fig_p008_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Configuration-selection procedure used to identify a working web-browsing setup for each AI assistant. [PITH_FULL_IMAGE:figures/full_fig_p010_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: Perplexity response stating it was unable to retrieve the requested contents but showing them (17359427 was [PITH_FULL_IMAGE:figures/full_fig_p011_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: Passive monitoring time series showing follow-up accesses to the controlled website outside the active testing [PITH_FULL_IMAGE:figures/full_fig_p016_5.png] view at source ↗

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

89 extracted references · 5 canonical work pages · 3 internal anchors

  1. [1]

    Greenwade

    George D. Greenwade. The C omprehensive T ex A rchive N etwork ( CTAN ). TUGBoat. 1993

    1993

  2. [2]

    Verify requests from Google crawlers and fetchers , year =

  3. [3]

    Business & Information Systems Engineering , volume=

    Foundation Models: A New Paradigm for Artificial Intelligence , author=. Business & Information Systems Engineering , volume=. 2024 , publisher=

    2024

  4. [4]

    GAIA: A Benchmark for General AI Assistants , booktitle =

    Mialon, Gr. GAIA: A Benchmark for General AI Assistants , booktitle =. 2024 , url =

    2024

  5. [5]

    Frontiers of Computer Science , volume=

    A survey on large language model based autonomous agents , author=. Frontiers of Computer Science , volume=. 2024 , publisher=. doi:10.1007/s11704-024-40231-1 , pages=

    doi:10.1007/s11704-024-40231-1 2024

  6. [6]

    Introducing the Search Arena: Evaluating Search-Enabled AI , year =

  7. [7]

    and Darrell, Trevor and Norouzi, Narges and Gonzalez, Joseph E

    Miroyan, Mihran and Wu, Tsung-Han and King, Logan and Li, Tianle and Pan, Jiayi and Hu, Xinyan and Chiang, Wei-Lin and Angelopoulos, Anastasios N. and Darrell, Trevor and Norouzi, Narges and Gonzalez, Joseph E. , title =. arXiv preprint arXiv:2506.05334 , year =

    arXiv

  8. [8]

    Computer Science Review , volume =

    A comprehensive survey of prompt engineering and context engineering techniques in large language models , author =. Computer Science Review , volume =. 2026 , issn =. doi:10.1016/j.cosrev.2026.100979 , url =

    arXiv 2026

  9. [9]

    Applied Sciences , volume=

    A Review of Current Trends, Techniques, and Challenges in Large Language Models (LLMs) , author=. Applied Sciences , volume=. 2024 , publisher=

    2024

  10. [10]

    Retrieval-Augmented Generation for Knowledge-Intensive NLP Tasks , booktitle =

    Lewis, Patrick and Perez, Ethan and Piktus, Aleksandra and Petroni, Fabio and Karpukhin, Vladimir and Goyal, Naman and K. Retrieval-Augmented Generation for Knowledge-Intensive NLP Tasks , booktitle =. 2020 , publisher =

    2020

  11. [11]

    2021 , howpublished =

    Nakano, Reiichiro and Hilton, Jacob and Balaji, Suchir and Wu, Jeff and Ouyang, Long and Kim, Christina and Hesse, Christopher and Jain, Shantanu and Kosaraju, Vineet and Saunders, William and Jiang, Xu and Cobbe, Karl and Eloundou, Tyna and Krueger, Gretchen and Button, Kevin and Knight, Matthew and Chess, Benjamin and Schulman, John , title =. 2021 , ho...

    2021

  12. [12]

    Grounding with Google Search , year =

  13. [13]

    URL context , year =

  14. [14]

    Agentic retrieval in Azure AI Search , year =

  15. [15]

    Regulation (EU) 2024/1689 of the European Parliament and of the Council of 13 June 2024 laying down harmonised rules on artificial intelligence and amending Regulations (EC) No 300/2008, (EU) No 167/2013, (EU) No 168/2013, (EU) 2018/858, (EU) 2018/1139 and (EU) 2019/2144 and Directives 2014/90/EU, (EU) 2016/797 and (EU) 2020/1828 (Artificial Intelligence ...

    2024

  16. [16]

    General-Purpose AI Models in the AI Act -- Questions & Answers , year =

  17. [17]

    AI Literacy -- Questions & Answers , year =

  18. [18]

    COMPL-AI Framework: A Technical Interpretation and LLM Benchmarking Suite for the EU Artificial Intelligence Act , journal =

    Guldimann, Philipp and Spiridonov, Alexander and Staab, Robin and Jovanovi. COMPL-AI Framework: A Technical Interpretation and LLM Benchmarking Suite for the EU Artificial Intelligence Act , journal =. 2024 , eprint =. doi:10.48550/arXiv.2410.07959 , url =

    Pith/arXiv arXiv doi:10.48550/arxiv.2410.07959 2024

  19. [19]

    TDM Reservation Protocol (TDMRep) , year =

  20. [20]

    Google's common crawlers , year =

  21. [21]

    2022 , month = sep, url =

    Koster, Martijn and Illyes, Gergely and Zeller, Henner and Sassman, Lucas , title =. 2022 , month = sep, url =

    2022

  22. [22]

    Overview of OpenAI Crawlers , year =

  23. [23]

    Overview of Google crawlers and fetchers (user agents) , year =

  24. [24]

    LMArena Search Arena Leaderboard , year =

  25. [25]

    2026 , howpublished =

    2026

  26. [26]

    ChatGPT , howpublished =

  27. [27]

    Claude , howpublished =

  28. [28]

    Copilot , howpublished =

  29. [29]

    DeepSeek Chat , howpublished =

  30. [30]

    Diffy Chat , howpublished =

  31. [31]

    Gemini , howpublished =

  32. [32]

    Grok , howpublished =

  33. [33]

    Mistral Chat , howpublished =

  34. [34]

    Perplexity , howpublished =

  35. [35]

    Qwen , howpublished =

  36. [36]

    NEXT-EVAL: Next Evaluation of Traditional and LLM Web Data Record Extraction

    Kim, Soyeon and Kim, Namhee and Jeong, Yeonwoo , title =. arXiv preprint arXiv:2505.17125 , year =. doi:10.48550/arXiv.2505.17125 , url =

    work page internal anchor Pith review Pith/arXiv arXiv doi:10.48550/arxiv.2505.17125

  37. [37]

    Data Science and Management , year =

    Xu, Zongben and Liu, Shan and Huang, Wei and Shi, Junmin and Xu, Fengmin and Tang, Xin and Lu, Haibing , title =. Data Science and Management , year =. doi:10.1016/j.dsm.2025.09.001 , url =

    work page doi:10.1016/j.dsm.2025.09.001 2025

  38. [38]

    AI in the Enterprise: How People Use M365 Copilot Chat

    Counts, Scott and Chen, Yan and Dong, Jing and Sharma, Himanshu and Zaikin, Andrey and Hu, Rui and Kok, Alperen and Yilmaz, Gorkem Ozer and Suri, Siddharth and Tomlinson, Kiran and Jaffe, Sonia and Wang, Will , title =. arXiv preprint arXiv:2605.23958 , year =. doi:10.48550/arXiv.2605.23958 , url =. 2605.23958 , archivePrefix =

    work page internal anchor Pith review Pith/arXiv arXiv doi:10.48550/arxiv.2605.23958

  39. [39]

    arXiv preprint arXiv:2604.07190 , year =

    The ATOM Report: Measuring the Open Language Model Ecosystem , author =. arXiv preprint arXiv:2604.07190 , year =. doi:10.48550/arXiv.2604.07190 , url =. 2604.07190 , archivePrefix =

    Pith/arXiv arXiv doi:10.48550/arxiv.2604.07190

  40. [40]

    Transactions on Machine Learning Research , issn=

    BrowserAgent: Building Web Agents with Human-Inspired Web Browsing Actions , author=. Transactions on Machine Learning Research , issn=. 2026 , url=

    2026

  41. [41]

    The Twelfth International Conference on Learning Representations , year =

    AgentBench: Evaluating LLMs as Agents , author=. The Twelfth International Conference on Learning Representations , year =

  42. [42]

    Qin, Yujia and Liang, Shihao and Ye, Yining and Zhu, Kunlun and Yan, Lan and Lu, Yaxi and Lin, Yankai and Cong, Xin and Tang, Xiangru and Qian, Bill and Zhao, Sihan and Hong, Lauren and Tian, Runchu and Xie, Ruobing and Zhou, Jie and Gerstein, Mark and Li, Dahai and Liu, Zhiyuan and Sun, Maosong , booktitle =. Tool. 2024 , url =

    2024

  43. [43]

    2025 , url =

    Yao, Shunyu and Shinn, Noah and Razavi, Pedram and Narasimhan, Karthik , booktitle =. 2025 , url =

    2025

  44. [44]

    AgentBoard: An Analytical Evaluation Board of Multi-turn LLM Agents , url =

    Ma, Chang and Zhang, Junlei and Zhu, Zhihao and Yang, Cheng and Yang, Yujiu and Jin, Yaohui and Lan, Zhenzhong and Kong, Lingpeng and He, Junxian , booktitle =. AgentBoard: An Analytical Evaluation Board of Multi-turn LLM Agents , url =. doi:10.52202/079017-2365 , editor =

    doi:10.52202/079017-2365

  45. [45]

    A Survey on Evaluation of LLM -based Agents

    Yehudai, Asaf and Eden, Lilach and Li, Alan and Uziel, Guy and Zhao, Yilun and Bar-Haim, Roy and Cohan, Arman and Shmueli-Scheuer, Michal. A Survey on Evaluation of LLM -based Agents. Findings of the A ssociation for C omputational L inguistics: ACL 2026. 2026. doi:10.18653/v1/2026.findings-acl.1330

    doi:10.18653/v1/2026.findings-acl.1330 2026

  46. [46]

    2025 , eprint =

    Multi-Agent Risks from Advanced AI , author =. 2025 , eprint =

    2025

  47. [47]

    Advances in Neural Information Processing Systems , editor =

    Deng, Xiang and Gu, Yu and Zheng, Boyuan and Chen, Shijie and Stevens, Sam and Wang, Boshi and Sun, Huan and Su, Yu , title =. Advances in Neural Information Processing Systems , editor =. 2023 , url =

    2023

  48. [48]

    WebArena: A Realistic Web Environment for Building Autonomous Agents , url =

    Zhou, Shuyan and Xu, Frank F and Zhu, Hao and Zhou, Xuhui and Lo, Robert and Sridhar, Abishek and Cheng, Xianyi and Ou, Tianyue and Bisk, Yonatan and Fried, Daniel and Alon, Uri and Neubig, Graham , booktitle =. WebArena: A Realistic Web Environment for Building Autonomous Agents , url =

  49. [49]

    and Verme, Manuel Del and Marty, Tom and Vazquez, David and Chapados, Nicolas and Lacoste, Alexandre , title =

    Drouin, Alexandre and Gasse, Maxime and Caccia, Massimo and Laradji, Issam H. and Verme, Manuel Del and Marty, Tom and Vazquez, David and Chapados, Nicolas and Lacoste, Alexandre , title =. Proceedings of the 41st International Conference on Machine Learning , articleno =. 2024 , publisher =

    2024

  50. [50]

    WorkArena++: Towards Compositional Planning and Reasoning-based Common Knowledge Work Tasks , url =

    Boisvert, L\'. WorkArena++: Towards Compositional Planning and Reasoning-based Common Knowledge Work Tasks , url =. Advances in Neural Information Processing Systems , doi =

  51. [51]

    and Zhou, Shuyan and Neubig, Graham

    Song, Yueqi and Xu, Frank F. and Zhou, Shuyan and Neubig, Graham. Beyond Browsing: API -Based Web Agents. Findings of the Association for Computational Linguistics: ACL 2025. 2025. doi:10.18653/v1/2025.findings-acl.577

    doi:10.18653/v1/2025.findings-acl.577 2025

  52. [52]

    arXiv preprint arXiv:2504.12516 , year =

    Wei, Jason and Sun, Zhiqing and Papay, Spencer and McKinney, Scott and Han, Jeffrey and Fulford, Isa and Chung, Hyung Won and Tachard Passos, Alex and Fedus, William and Glaese, Amelia , title =. arXiv preprint arXiv:2504.12516 , year =. doi:10.48550/arXiv.2504.12516 , url =. 2504.12516 , archivePrefix =

    Pith/arXiv arXiv doi:10.48550/arxiv.2504.12516

  53. [53]

    A Survey of Web Content Control for Generative AI , booktitle =

    Dinzinger, Michael and He. A Survey of Web Content Control for Generative AI , booktitle =. 2024 , month = mar, publisher =

    2024

  54. [54]

    2025 , isbn =

    Kim, Taein and Bock, Karstan and Luo, Claire and Liswood, Amanda and Poroslay, Chloe and Wenger, Emily , title =. 2025 , isbn =. doi:10.1145/3730567.3764471 , booktitle =

    arXiv 2025

  55. [55]

    and Zhao, Ben Y

    Liu, Enze and Luo, Elisa and Shan, Shawn and Voelker, Geoffrey M. and Zhao, Ben Y. and Savage, Stefan , title =. 2025 , isbn =. doi:10.1145/3730567.3732913 , booktitle =

    arXiv 2025

  56. [56]

    ai.txt: A Domain-Specific Language for Guiding AI Interactions with the Internet

    Li, Yuekang and Song, Wei and Zhu, Bangshuo and Gong, Dong and Liu, Yi and Deng, Gelei and Chen, Chunyang and Ma, Lei and Sun, Jun and Walsh, Toby and Xue, Jingling , title =. arXiv preprint arXiv:2505.07834 , year =. doi:10.48550/arXiv.2505.07834 , url =. 2505.07834 , archivePrefix =

    work page internal anchor Pith review Pith/arXiv arXiv doi:10.48550/arxiv.2505.07834

  57. [57]

    2026 , isbn =

    Steinacker-Olsztyn, Nicolas and Gosain, Devashish and Dao, Ha , title =. 2026 , isbn =. doi:10.1145/3774904.3792625 , booktitle =

    arXiv 2026

  58. [58]

    and Geng, Longling , title =

    Chang, Edward Y. and Geng, Longling , title =. Proc. VLDB Endow. , month = aug, pages =. 2025 , issue_date =. doi:10.14778/3750601.3750611 , abstract =

    arXiv 2025

  59. [59]

    and Wu, Ziyang and Zhang, Tianjun and Keutzer, Kurt and Gonzalez, Joseph E

    Tan, Sijun and Li, Xiuyu and Patil, Shishir G. and Wu, Ziyang and Zhang, Tianjun and Keutzer, Kurt and Gonzalez, Joseph E. and Popa, Raluca Ada , title =. Proceedings of the 2024 Conference on Empirical Methods in Natural Language Processing , year =. doi:10.18653/v1/2024.emnlp-main.975 , url =

    doi:10.18653/v1/2024.emnlp-main.975 2024

  60. [60]

    Carl , title =

    Drott, M. Carl , title =. Information Processing & Management , year =. doi:10.1016/S0306-4573(01)00039-5 , url =

    work page doi:10.1016/s0306-4573(01)00039-5

  61. [61]

    Information Processing & Management , year =

    Sun, Shiqi and Zhang, Kun and Li, Jingyuan and Yu, Min and Hou, Kun and Wang, Yuanzhuo and Cheng, Xueqi , title =. Information Processing & Management , year =. doi:10.1016/j.ipm.2025.104147 , url =

    arXiv 2025

  62. [62]

    QwenLong-CPRS: Towards

    Shen, Weizhou and Li, Chenliang and Wan, Fanqi and Liao, Shengyi and Lai, Shaopeng and Zhang, Bo and Shi, Yingcheng and Wu, Yuning and Fu, Gang and Li, Zhansheng and Yang, Bin and Zhang, Ji and Huang, Fei and Zhou, Jingren and Yan, Ming , journal =. QwenLong-CPRS: Towards. 2025 , eprint =. doi:10.48550/arXiv.2505.18092 , url =

    Pith/arXiv arXiv doi:10.48550/arxiv.2505.18092 2025

  63. [63]

    arXiv preprint arXiv:2603.09619 , year =

    Context Engineering: From Prompts to Corporate Multi-Agent Architecture , author =. arXiv preprint arXiv:2603.09619 , year =. doi:10.48550/arXiv.2603.09619 , url =. 2603.09619 , archivePrefix =

    doi:10.48550/arxiv.2603.09619

  64. [64]

    Effective context engineering for AI agents , year =

  65. [65]

    2024 , month = oct, howpublished =

    Introducing. 2024 , month = oct, howpublished =

    2024

  66. [66]

    2026 , month = jun, url =

    Gottfried, Jeffrey and Bishop, William and Anderson, Monica and Faverio, Michelle and Park, Eugenie and McClain, Colleen , title =. 2026 , month = jun, url =

    2026

  67. [67]

    2023 , month = jul, doi =

    Xu, Ruiyun and Feng, Yue (Katherine) and Chen, Hailiang , title =. 2023 , month = jul, doi =

    2023

  68. [68]

    2024 , month = jul, howpublished =

    Fishkin, Rand , title =. 2024 , month = jul, howpublished =

    2024

  69. [69]

    2026 , month = jun, howpublished =

    Fishkin, Rand , title =. 2026 , month = jun, howpublished =

    2026

  70. [70]

    2025 , month = mar, howpublished =

    Ja. 2025 , month = mar, howpublished =

    2025

  71. [71]

    2026 , month = feb, howpublished =

    European Publishers Council Files Formal Antitrust Complaint against Google over. 2026 , month = feb, howpublished =

    2026

  72. [72]

    2019 , howpublished =

    Directive (. 2019 , howpublished =

    2019

  73. [73]

    Code of Practice for General-Purpose

    Peukert, Alexander and Castets-Renard, C. Code of Practice for General-Purpose. 2025 , month = jul, url =

    2025

  74. [74]

    How Your Data Is Used to Improve Model Performance , year =

  75. [75]

    Is My Data Used for Model Training? , year =

  76. [76]

    Microsoft Copilot Privacy Controls , year =

  77. [77]

    2025 , month = may, howpublished =

    2025

  78. [78]

    Do You Use My User Data to Train Your Artificial Intelligence Models? , year =

  79. [79]

    Data Collection at Perplexity , year =

  80. [80]

    2026 , month = feb, howpublished =

    2026

Showing first 80 references.