pith. sign in

arxiv: 2605.23676 · v1 · pith:LDSBI54Gnew · submitted 2026-05-22 · 💻 cs.HC

AI at the Front Lines of Platform Governance: Using LLMs to Support Illegal Content Reporting under the Digital Services Act

Marie-Therese Sekwenz , Shreyan Biswas , Rita Hermann-Gsenger , Ujwal Gadiraju This is my paper

Pith reviewed 2026-05-25 03:20 UTC · model grok-4.3

classification 💻 cs.HC
keywords LLM assistanceillegal content reportingDigital Services Actuser studyexplainable AIevaluative AIcontent moderationplatform governance
0
0 comments X

The pith

An AI assistant presenting balanced arguments for legal categories improves user accuracy in DSA illegal content reports when the AI errs, unlike single-suggestion XAI.

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

The paper examines how LLM assistants can help users create sufficiently substantiated notices of illegal online content under EU Digital Services Act Article 16. It compares unaided reporting against a conventional XAI that offers one category with rationale and an EvalAI that supplies pro and con arguments for multiple options, all tested under controlled AI error conditions in a 450-participant study. EvalAI raised provision-level accuracy and lowered misclassification distance relative to XAI, especially on near-miss and overbreadth mistakes. Conventional XAI produced quicker decisions only when the AI output was already correct, yet neither assistant improved the quality of users' written explanations over unaided work. The work identifies design trade-offs for compliance interfaces that must handle imperfect AI while supporting good-faith reporting.

Core claim

In a controlled user study using an interface modeled on a major platform reporting workflow, the evaluative AI assistant that presents balanced pro and con arguments across candidate legal provisions improves provision-level accuracy under AI error and reduces misclassification distance relative to conventional XAI, particularly for near-miss and overbreadth errors; when AI output is correct, conventional XAI enables faster decisions, but neither AI assistance form reliably improves the quality of users' substantiated explanations relative to unaided reporting.

What carries the argument

EvalAI, the evaluative LLM assistant that presents balanced pro and con arguments across candidate legal provisions instead of a single suggested category.

If this is right

  • Platforms adopting EvalAI-style interfaces may observe fewer misclassifications in user notices even when the underlying LLM produces errors.
  • Conventional single-suggestion XAI may be suitable only in settings where the AI is known to be highly accurate, to preserve decision speed.
  • Neither form of AI assistance improves the substantive quality of user explanations, indicating that additional interface elements are needed to meet the good-faith substantiation requirement.
  • Design choices for DSA reporting tools must explicitly weigh gains in accuracy against risks of over-reliance on flawed AI output.

Where Pith is reading between the lines

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

  • The balanced-argument approach could apply to other regulatory tasks where users must translate ambiguous facts into fixed legal categories.
  • Interfaces might add explicit prompts about common AI error patterns to further reduce the distance of remaining misclassifications.
  • Live platform trials would be required to check whether the controlled error regimes match the distribution of mistakes produced by actual deployed LLMs.

Load-bearing premise

The simulated AI error regimes and the modeled reporting interface accurately reflect real LLM behavior and actual user decisions when submitting DSA Article 16 notices.

What would settle it

A deployment on a live platform that measures actual user reports against final legal determinations under real LLM outputs would show whether the accuracy gains and error reductions observed in the controlled study appear in practice.

Figures

Figures reproduced from arXiv: 2605.23676 by Marie-Therese Sekwenz, Rita Hermann-Gsenger, Shreyan Biswas, Ujwal Gadiraju.

Figure 1
Figure 1. Figure 1: Components of the experimental interface. (a) the social media feed the users were presented [PITH_FULL_IMAGE:figures/full_fig_p007_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Overview of the experimental flow, assistance conditions, and randomized AI error conditions used in the study. [PITH_FULL_IMAGE:figures/full_fig_p009_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Provision-level accuracy under AI error regimes. (a) Overall accuracy on AI-error trials by assistance condition. (b) [PITH_FULL_IMAGE:figures/full_fig_p011_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: Misclassification distance under AI error by error condition and assistance condition. [PITH_FULL_IMAGE:figures/full_fig_p011_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: Downstream enforcement risk and explanation quality. [PITH_FULL_IMAGE:figures/full_fig_p013_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: Decision time on no-error trials by assistance condition. [PITH_FULL_IMAGE:figures/full_fig_p022_6.png] view at source ↗
Figure 7
Figure 7. Figure 7: Example of Facebook’s illegal content reporting UI for Germany [PITH_FULL_IMAGE:figures/full_fig_p030_7.png] view at source ↗
read the original abstract

Illegal content reporting mechanisms are a key technical and organizational measure through which online platforms address illegal content under the European Union Digital Services Act (DSA). Article 16 requires user notices to be sufficiently substantiated and submitted in good faith, placing users in the difficult position of interpreting legal and procedural language and translating ambiguous content into legally meaningful categories and reasons. We investigate how large language model (LLM)-based assistants can support this reporting process. In a controlled user study (N = 450) using an interface modeled on a major platform reporting workflow, we compare three conditions: unaided reporting, a conventional explainable AI assistant (XAI) that suggests a single legal category with a rationale, and an evaluative AI assistant (EvalAI) that presents balanced pro and con arguments across candidate legal provisions. We further examine these assistance forms under systematically varied AI error regimes. Our results show that EvalAI improves provision-level accuracy under AI error and reduces misclassification distance relative to conventional XAI, particularly for near-miss and overbreadth errors. When AI output is correct, conventional XAI enables faster decisions, but neither AI assistance form reliably improves the quality of users' substantiated explanations relative to unaided reporting. We discuss design implications for compliance-oriented reporting interfaces, highlighting trade-offs between accuracy, deliberation, explanation quality, and vulnerability to misleading AI output.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

3 major / 0 minor

Summary. The paper reports a controlled user study (N=450) comparing unaided reporting, conventional XAI (single category + rationale), and EvalAI (balanced pro/con arguments) for DSA Article 16 illegal content notices. Using an interface modeled on platform workflows and systematically varied AI error regimes, it claims EvalAI improves provision-level accuracy and reduces misclassification distance under AI error (especially near-miss and overbreadth), conventional XAI speeds decisions when correct, and neither assistance reliably improves substantiated explanation quality over unaided reporting. Design implications for compliance interfaces are discussed.

Significance. If the results hold, the work supplies empirical evidence on trade-offs in AI assistance for legal classification tasks under the DSA, showing that evaluative (pro/con) formats can mitigate certain AI errors better than conventional XAI while highlighting limits on explanation quality. It contributes to HCI and platform governance by testing concrete interface designs against realistic error conditions and offering actionable recommendations for reporting tools.

major comments (3)
  1. [Abstract / Study Design] Abstract / Study Design: The headline result (EvalAI improves provision-level accuracy and reduces misclassification distance under AI error) depends on the specific error regimes (near-miss, overbreadth). The description provides no indication that these regimes were sampled from or validated against observed LLM outputs on the same DSA Article 16 items; if constructed by experimenters, the measured advantage of the pro/con format could be an artifact of error selection rather than a robust property of EvalAI.
  2. [Abstract] Abstract: The study is described as controlled with N=450 and systematic error regimes, yet the abstract supplies no statistical details, effect sizes, exclusion criteria, power analysis, or raw data summaries. This prevents verification of the central accuracy and misclassification-distance claims from the available text.
  3. [Abstract / Study Design] Abstract / Study Design: The weakest assumption is that the simulated AI error regimes and the modeled interface accurately capture real-world LLM behavior and user decision processes in DSA Article 16 tasks. No evidence is given that the error distributions match actual LLM failure modes on legal classification, which directly affects the generalizability of the EvalAI advantage.

Simulated Author's Rebuttal

3 responses · 0 unresolved

We thank the referee for their constructive feedback. We address each major comment below and note planned revisions.

read point-by-point responses

  1. Referee: [Abstract / Study Design] Abstract / Study Design: The headline result (EvalAI improves provision-level accuracy and reduces misclassification distance under AI error) depends on the specific error regimes (near-miss, overbreadth). The description provides no indication that these regimes were sampled from or validated against observed LLM outputs on the same DSA Article 16 items; if constructed by experimenters, the measured advantage of the pro/con format could be an artifact of error selection rather than a robust property of EvalAI.

    Authors: The error regimes were constructed to represent common LLM failure modes in legal classification (near-miss and overbreadth) identified via pilot testing on similar DSA items. We did not perform systematic sampling or validation of distributions against LLM outputs on the exact study items. We will revise the methods and discussion to detail the construction process and explicitly note this as a limitation on the robustness of the EvalAI advantage. revision: partial

  2. Referee: [Abstract] Abstract: The study is described as controlled with N=450 and systematic error regimes, yet the abstract supplies no statistical details, effect sizes, exclusion criteria, power analysis, or raw data summaries. This prevents verification of the central accuracy and misclassification-distance claims from the available text.

    Authors: We agree the abstract should include more statistical information for transparency. The revised abstract will report key effect sizes, p-values for main accuracy and misclassification findings, and a brief note on sample and analysis. revision: yes

  3. Referee: [Abstract / Study Design] Abstract / Study Design: The weakest assumption is that the simulated AI error regimes and the modeled interface accurately capture real-world LLM behavior and user decision processes in DSA Article 16 tasks. No evidence is given that the error distributions match actual LLM failure modes on legal classification, which directly affects the generalizability of the EvalAI advantage.

    Authors: We acknowledge the external validity concern. The regimes were designed as controlled tests of assistance formats under representative error types, and the interface was modeled on actual platform workflows. We will expand the limitations and future work sections to discuss the assumptions about real-world LLM distributions and user processes, and suggest validation approaches. revision: partial

Circularity Check

0 steps flagged

Empirical user study with direct experimental comparisons; no derivation chain or fitted quantities

full rationale

The paper reports results from a controlled user study (N=450) comparing three reporting conditions under varied AI error regimes. All load-bearing claims are direct empirical contrasts (accuracy, decision time, explanation quality) between unaided, XAI, and EvalAI arms. No equations, parameter fitting, self-definitional constructs, or self-citation chains appear in the derivation of the headline results. The study design is self-contained; measured outcomes are not redefined in terms of the inputs or prior outputs.

Axiom & Free-Parameter Ledger

0 free parameters · 2 axioms · 0 invented entities

Empirical HCI study relying on standard experimental assumptions rather than new mathematical constructs or postulated entities.

axioms (2)
  • domain assumption Random assignment of participants to conditions produces comparable groups without systematic bias
    Standard assumption in controlled user studies described in the abstract.
  • domain assumption The legal categories and error regimes used are representative of actual DSA Article 16 requirements and LLM outputs
    Invoked by the study design that models a major platform workflow.

pith-pipeline@v0.9.0 · 5794 in / 1396 out tokens · 35071 ms · 2026-05-25T03:20:19.215360+00:00 · methodology

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

136 extracted references · 136 canonical work pages · 2 internal anchors

  1. [1]

    AG Bonn. 2016. Urteil vom 10.08.2016. https://openjur.de/u/2154100.html

    work page arXiv 2016

  2. [2]

    AG Hannover. 2020. Urteil vom 03.02.2020. https://openjur.de/u/2316520.html FAccT ’26, June 25–28, 2026, Montreal, QC, Canada Marie-Therese Sekwenz, Shreyan Biswas, Rita Hermann-Gsenger, and Ujwal Gadiraju

    work page arXiv 2020

  3. [3]

    Garrett Allen, Mike Beijen, David Maxwell, and Ujwal Gadiraju. 2023. In a Hurry: How Time Constraints and the Presentation of Web Search Results Affect User Behaviour and Experience. InWeb Engineering: 23rd International Conference, ICWE 2023, Alicante, Spain, June 6–9, 2023, Proceedings. Springer-Verlag, Berlin, Heidelberg, 221–235. doi:10.1007/978-3-031...

    work page doi:10.1007/978-3-031-34444-2_16 2023

  4. [4]

    Marco Aspromonte, Andrea Ferraris, Federico Galli, and Giuseppe Contissa. 2024. LLMs to the Rescue: Explaining DSA Statements of Reason with Platform’s Terms of Services. InProceedings of the Natural Legal Language Processing Workshop 2024(Miami, FL, USA, 2024). Association for Computational Linguistics, Miami, FL, USA, 205–215. doi:10.18653/v1/2024.nllp-1.17

    work page doi:10.18653/v1/2024.nllp-1.17 2024

  5. [5]

    Solon Barocas, Moritz Hardt, and Arvind Narayanan. 2020. Fairness and machine learning.Recommender systems handbook1 (2020), 453–459

    work page 2020

  6. [6]

    Bayern OLG. 2020. Beschluss vom 14.02.2020. https://openjur.de/u/2274641.html

    work page arXiv 2020

  7. [7]

    Eagan, and Winston Maxwell

    Astrid Bertrand, Rafik Belloum, James R. Eagan, and Winston Maxwell. 2022. How Cognitive Biases Affect XAI-assisted Decision- making: A Systematic Review. InProceedings of the 2022 AAAI/ACM Conference on AI, Ethics, and Society (AIES ’22). Association for Computing Machinery, New York, NY, USA, 78–91. doi:10.1145/3514094.3534164

    work page doi:10.1145/3514094.3534164 2022

  8. [8]

    BGH. 2013. Beschluss vom 12.11.2013. https://openjur.de/u/662251.html

    work page 2013

  9. [9]

    BGH. 2017. Beschluss vom 01.02.2017. https://openjur.de/u/2116511.html

    work page arXiv 2017

  10. [10]

    BGH. 2017. Beschluss vom 07.09.2017. https://openjur.de/u/2110510.html

    work page arXiv 2017

  11. [12]

    BGH. 2018. Beschluss vom 19.06.2018. https://openjur.de/u/2121414.html

    work page arXiv 2018

  12. [13]

    BGH. 2018. Beschluss vom 22.03.2018. https://openjur.de/u/2110505.html

    work page arXiv 2018

  13. [14]

    BGH. 2021. Beschluss vom 11.08.2021. https://openjur.de/u/2383741.html

    work page arXiv 2021

  14. [15]

    Shreyan Biswas, Alexander Erlei, and Ujwal Gadiraju. 2025. Mind the Gap! Choice Independence in Using Multilingual LLMs for Persuasive Co-Writing Tasks in Different Languages. InProceedings of the 2025 CHI Conference on Human Factors in Computing Systems (CHI ’25). Association for Computing Machinery, New York, NY, USA, 1–20. doi:10.1145/3706598.3713201

    work page doi:10.1145/3706598.3713201 2025

  15. [16]

    Shreyan Biswas, Alexander Erlei, and Ujwal Gadiraju. 2026. Belief Updating and Delegation in Multi-Task Human–AI Interaction: Evidence from Controlled Simulations. InProceedings of the 2026 CHI Conference on Human Factors in Computing Systems. 1–23

    work page 2026

  16. [17]

    Zana Buçinca, Maja Barbara Malaya, and Krzysztof Z Gajos. 2021. To trust or to think: cognitive forcing functions can reduce overreliance on AI in AI-assisted decision-making.Proceedings of the ACM on Human-computer Interaction5, CSCW1 (2021), 1–21

    work page 2021

  17. [18]

    Koch, Samantha Straka, Marc Erich Latoschik, and Carolin Wienrich

    Astrid Carolus, Martin J. Koch, Samantha Straka, Marc Erich Latoschik, and Carolin Wienrich. 2023. MAILS - Meta AI literacy scale: Development and testing of an AI literacy questionnaire based on well-founded competency models and psychological change- and meta-competencies.Computers in Human Behavior: Artificial Humans1, 2 (2023), 100014. doi:10.1016/j.c...

    work page doi:10.1016/j.chbah.2023.100014 2023

  18. [19]

    Caroline Cauffman and Catalina Goanta. 2021. A New Order: The Digital Services Act and Consumer Protection.European Journal of Risk Regulation12, 4 (Dec. 2021), 758–774. doi:10.1017/err.2021.8

    work page doi:10.1017/err.2021.8 2021

  19. [20]

    Edward Y. Chang. 2024. SocraSynth: Multi-LLM Reasoning with Conditional Statistics. doi:10.48550/arXiv.2402.06634 arXiv:2402.06634 [cs]

    work page doi:10.48550/arxiv.2402.06634 2024

  20. [21]

    Edward Y. Chang. 2024. SocraSynth: Multi-LLM Reasoning with Conditional Statistics. arXiv:2402.06634 [cs.AI] https://arxiv.org/abs/ 2402.06634

    work page arXiv 2024

  21. [22]

    Simon Chesterman. 2024. Lawful but Awful: Evolving Legislative Responses to Address Online Misinformation, Disinformation, and Mal-Information in the Age of Generative AI.The American Journal of Comparative Law72, 4 (2024). doi:10.1093/ajcl/avaf020

    work page doi:10.1093/ajcl/avaf020 2024

  22. [23]

    Conor Clune and Emma McDaid. 2024. Content moderation on social media: constructing accountability in the digital space.Accounting, Auditing & Accountability Journal37, 1 (Jan. 2024), 257–279. doi:10.1108/AAAJ-11-2022-6119

    work page doi:10.1108/aaaj-11-2022-6119 2024

  23. [24]

    Jennifer Cobbe, Michelle S. A. Lee, and Jatinder Singh. 2021. Reviewable Automated Decision-Making: A Framework for Accountable Algorithmic Systems. InProceedings of the 2021 ACM Conference on Fairness, Accountability, and Transparency (FAccT). ACM, New York, NY, USA, 1–14. doi:10.1145/3442188.3445921

    work page doi:10.1145/3442188.3445921 2021

  24. [25]

    European Commission. 2020. COMMISSION STAFF WORKING DOCUMENT IMPACT ASSESSMENT REPORT Accompanying the document Proposal for a REGULATION OF THE EUROPEAN PARLIAMENT AND OF THE COUNCIL on contestable and fair markets in the digital sector (Digital Markets Act). https://eur-lex.europa.eu/legal-content/EN/TXT/?uri=celex:52020SC0363

    work page 2020

  25. [26]

    European Commission. 2022. Regulation (EU) 2022/2065 of the European Parliament and of the Council of 19 October 2022 on a Single Market For Digital Services and amending Directive 2000/31/EC (Digital Services Act) (Text with EEA relevance). http: //data.europa.eu/eli/reg/2022/2065/oj/eng Legislative Body: EP, CONSIL

    work page 2022

  26. [27]

    European Commission. 2025. Commission fines X€120 million under the Digital Services Act. https://ec.europa.eu/commission/ presscorner/detail/en/ip_25_2934

    work page 2025

  27. [28]

    Court of Justice of the European Union. 2019. Fashion ID GmbH & CoKG v Verbraucherzentrale NRW eV. https://eur-lex.europa.eu/legal- content/en/TXT/?uri=CELEX:62017CJ0040

    work page 2019

  28. [29]

    Fred D. Davis. 1989. Perceived Usefulness, Perceived Ease of Use, and User Acceptance of Information Technology.MIS Quarterly13, 3 (1989), 319. doi:10.2307/249008 Supporting Illegal Content Reporting under the Digital Services Act with LLMs FAccT ’26, June 25–28, 2026, Montreal, QC, Canada

    work page doi:10.2307/249008 1989

  29. [30]

    Yang Deng, Lizi Liao, Wenqiang Lei, Grace Hui Yang, Wai Lam, and Tat-Seng Chua. 2025. Proactive Conversational AI: A Comprehensive Survey of Advancements and Opportunities.ACM Transactions on Information Systems43, 3 (2025), 1–45. doi:10.1145/3715097

    work page doi:10.1145/3715097 2025

  30. [31]

    Fabiana Di Porto and Alexander Egberts. 2023. The collective welfare dimension of dark patterns regulation.European Law Journal29, 1-2 (2023), 114–141. doi:10.1111/eulj.12478 _eprint: https://onlinelibrary.wiley.com/doi/pdf/10.1111/eulj.12478

    work page doi:10.1111/eulj.12478 2023

  31. [32]

    DiSalvo, Gabriela Viviana Saenz, W

    Lisa M. DiSalvo, Gabriela Viviana Saenz, W. Eric Wong, and Dongcheng Li. 2022. Social Media Safety Practices and Flagging Sensitive Posts. In2022 IEEE 22nd International Conference on Software Quality, Reliability, and Security Companion (QRS-C). 8–15. doi:10.1109/QRS-C57518.2022.00012

    work page doi:10.1109/qrs-c57518.2022.00012 2022

  32. [33]

    Evelyn Douek. 2022. Content Moderation as Systems Thinking.Harvard Law Review36, 2 (Dec. 2022), 528–606. https://harvardlawreview. org/print/vol-136/content-moderation-as-systems-thinking/

    work page 2022

  33. [34]

    Mary T Dzindolet, Linda G Pierce, Hall P Beck, and Lloyd A Dawe. 2002. The perceived utility of human and automated aids in a visual detection task.Human factors44, 1 (2002), 79–94

    work page 2002

  34. [35]

    Upol Ehsan, Q Vera Liao, Samir Passi, Mark O Riedl, and Hal Daumé III. 2024. Seamful xai: Operationalizing seamful design in explainable ai.Proceedings of the ACM on Human-Computer Interaction8, CSCW1 (2024), 1–29

    work page 2024

  35. [36]

    Mohamed Elaraby, Diane Litman, Xiang Lorraine Li, and Ahmed Magooda. 2024. Persuasiveness of Generated Free-Text Rationales in Subjective Decisions: A Case Study on Pairwise Argument Ranking. InFindings of the Association for Computational Linguistics: EMNLP 2024. Association for Computational Linguistics, Miami, Florida, USA, 14311–14329. doi:10.18653/v1...

    work page doi:10.18653/v1/2024.findings-emnlp.836 2024

  36. [37]

    Alexander Erlei, Federico Maria Cau, Radoslav Georgiev, Sagar Chethan Kumar, Kilian Bizer, and Ujwal Gadiraju. 2026. When Life Gives You AI, Will You Turn It Into A Market for Lemons? Understanding How Information Asymmetries About AI System Capabilities Affect Market Outcomes and Adoption. InProceedings of the 2026 CHI Conference on Human Factors in Comp...

    work page 2026

  37. [38]

    Alexander Erlei, Abhinav Sharma, and Ujwal Gadiraju. 2024. Understanding Choice Independence and Error Types in Human-AI Collaboration. InProceedings of the 2024 CHI Conference on Human Factors in Computing Systems (CHI ’24). Association for Computing Machinery, New York, NY, USA, 1–19. doi:10.1145/3613904.3641946

    work page doi:10.1145/3613904.3641946 2024

  38. [39]

    Raymond Fok and Daniel S Weld. 2024. In search of verifiability: Explanations rarely enable complementary performance in AI-advised decision making.AI Magazine45, 3 (2024), 317–332

    work page 2024

  39. [40]

    Gabriel Freedman, Adam Dejl, Deniz Gorur, Xiang Yin, Antonio Rago, and Francesca Toni. 2025. Argumentative Large Language Models for Explainable and Contestable Claim Verification. arXiv:2405.02079 [cs.CL] https://arxiv.org/abs/2405.02079

    work page internal anchor Pith review Pith/arXiv arXiv 2025

  40. [41]

    Ujwal Gadiraju and Agathe Balayn. 2025. The Enterprising and Elusive Prospects of Human-AI Collaboration. InEnterprise AI. Springer, 211–243

    work page 2025

  41. [42]

    2018.Custodians of the Internet: Platforms, content moderation, and the hidden decisions that shape social media

    Tarleton Gillespie. 2018.Custodians of the Internet: Platforms, content moderation, and the hidden decisions that shape social media. Yale University Press, New Haven, CT

    work page 2018

  42. [43]

    Tarleton Gillespie. 2020. Content moderation, AI, and the question of scale.Big Data & Society7, 2 (July 2020), 2053951720943234. doi:10.1177/2053951720943234 Publisher: SAGE Publications Ltd

    work page doi:10.1177/2053951720943234 2020

  43. [44]

    Juan Felipe Gomez, Caio Machado, Lucas Monteiro Paes, and Flavio Calmon. 2024. Algorithmic Arbitrariness in Content Moderation. In Proceedings of the 2024 ACM Conference on Fairness, Accountability, and Transparency (FAccT ’24). Association for Computing Machinery, New York, NY, USA, 2234–2253. doi:10.1145/3630106.3659036

    work page doi:10.1145/3630106.3659036 2024

  44. [45]

    Robert Gorwa, Reuben Binns, and Christian Katzenbach. 2020. Algorithmic content moderation: Technical and political challenges in the automation of platform governance.Big Data & Society7, 1 (2020), 205395171989794. doi:10.1177/2053951719897945

    work page doi:10.1177/2053951719897945 2020

  45. [46]

    Sreeprasad Govindankutty and Avneesh Kumar. 2024. Design and Implementation of Automated Content Moderation Systems in Social Media.Integrated Journal for Research in Arts and Humanities4, 6 (Nov. 2024), 380–402. doi:10.55544/ijrah.4.6.27

    work page doi:10.55544/ijrah.4.6.27 2024

  46. [47]

    Gray, Ritika Gairola, Nayah Boucaud, Maliha Hashmi, Shruthi Sai Chivukula, Ambika R Menon, and Ja-Nae Duane

    Colin M. Gray, Ritika Gairola, Nayah Boucaud, Maliha Hashmi, Shruthi Sai Chivukula, Ambika R Menon, and Ja-Nae Duane. 2024. Legal Trouble?: UX Practitioners’ Engagement with Law and Regulation. InCompanion Publication of the 2024 ACM Designing Interactive Systems Conference (DIS ’24 Companion). Association for Computing Machinery, New York, NY, USA, 106–1...

    work page doi:10.1145/3656156 2024

  47. [48]

    Gray, Yubo Kou, Bryan Battles, Joseph Hoggatt, and Austin L

    Colin M. Gray, Yubo Kou, Bryan Battles, Joseph Hoggatt, and Austin L. Toombs. 2018. The Dark (Patterns) Side of UX Design. In Proceedings of the 2018 CHI Conference on Human Factors in Computing Systems (CHI ’18). Association for Computing Machinery, New York, NY, USA, 1–14. doi:10.1145/3173574.3174108

    work page doi:10.1145/3173574.3174108 2018

  48. [49]

    Gray, Cristiana Teixeira Santos, Nataliia Bielova, and Thomas Mildner

    Colin M. Gray, Cristiana Teixeira Santos, Nataliia Bielova, and Thomas Mildner. 2024. An Ontology of Dark Patterns Knowledge: Foundations, Definitions, and a Pathway for Shared Knowledge-Building. InProceedings of the 2024 CHI Conference on Human Factors in Computing Systems (CHI ’24). Association for Computing Machinery, New York, NY, USA, 1–22. doi:10.1...

    work page doi:10.1145/3613904.3642436 2024

  49. [50]

    Akshit Gupta, Debadeep Basu, Ramya Ghantasala, Sihang Qiu, and Ujwal Gadiraju. 2022. To Trust or Not To Trust: How a Conversational Interface Affects Trust in a Decision Support System. InProceedings of the ACM Web Conference 2022 (WWW ’22). Association for Computing Machinery, New York, NY, USA, 3531–3540. doi:10.1145/3485447.3512248

    work page doi:10.1145/3485447.3512248 2022

  50. [51]

    Philipp Hacker, Ralf Krestel, Stefan Grundmann, and Felix Naumann. 2020. Explainable AI under Contract and Tort Law: Legal Incentives and Technical Challenges.Artificial Intelligence and Law28, 4 (2020), 415–439. doi:10.1007/s10506-020-09260-6 Published online 19 January 2020; print December 2020. FAccT ’26, June 25–28, 2026, Montreal, QC, Canada Marie-Th...

    work page doi:10.1007/s10506-020-09260-6 2020

  51. [52]

    Hart and Lowell E

    Sandra G. Hart and Lowell E. Staveland. 1988. Development of NASA-TLX (Task Load Index): Results of Empirical and Theoretical Research. InAdvances in Psychology. Vol. 52. Elsevier, Amsterdam, 139–183. doi:10.1016/S0166-4115(08)62386-9

    work page doi:10.1016/s0166-4115(08)62386-9 1988

  52. [53]

    Gaole He, Nilay Aishwarya, and Ujwal Gadiraju. 2025. Is Conversational XAI All You Need? Human-AI Decision Making With a Conversational XAI Assistant. InProceedings of the 30th International Conference on Intelligent User Interfaces (IUI ’25). Association for Computing Machinery, New York, NY, USA, 907–924. doi:10.1145/3708359.3712133

    work page doi:10.1145/3708359.3712133 2025

  53. [54]

    Gaole He, Agathe Balayn, Stefan Buijsman, Jie Yang, and Ujwal Gadiraju. 2024. Opening the Analogical Portal to Explainability: Can Analogies Help Laypeople in AI-assisted Decision Making?Journal of Artificial Intelligence Research81 (2024), 117–162

    work page 2024

  54. [55]

    Gaole He, Abri Bharos, and Ujwal Gadiraju. 2024. To err is AI! Debugging as an intervention to facilitate appropriate reliance on AI systems. InProceedings of the 35th ACM Conference on Hypertext and Social Media. 98–105

    work page 2024

  55. [56]

    Gaole He, Gianluca Demartini, and Ujwal Gadiraju. 2025. Plan-Then-Execute: An Empirical Study of User Trust and Team Performance When Using LLM Agents As A Daily Assistant. InProceedings of the 2025 CHI Conference on Human Factors in Computing Systems. Number 414 in CHI ’25. Association for Computing Machinery, New York, NY, USA, 1–22. https://doi.org/10....

    work page doi:10.1145/3706598.3713218 2025

  56. [57]

    Gaole He, Lucie Kuiper, and Ujwal Gadiraju. 2023. Knowing About Knowing: An Illusion of Human Competence Can Hinder Appropriate Reliance on AI Systems. InProceedings of the 2023 CHI Conference on Human Factors in Computing Systems (CHI ’23). Association for Computing Machinery, New York, NY, USA, 1–18. doi:10.1145/3544548.3581025

    work page doi:10.1145/3544548.3581025 2023

  57. [58]

    Johanna Herman. 2024. Dark patterns: EU’s regulatory efforts.SECURITY AND PRIV ACY7, 6 (2024), e441. doi:10.1002/spy2.441 _eprint: https://onlinelibrary.wiley.com/doi/pdf/10.1002/spy2.441

    work page doi:10.1002/spy2.441 2024

  58. [59]

    I thought it was my mistake, but it’s really the design

    Sudhamshu Hosamane, Alyvia Walters, Yao Lyu, and Shagun Jhaver. 2025. "I thought it was my mistake, but it’s really the design”: A Critical Examination of the Accessibility of User-Enacted Moderation Tools on Facebook and X. doi:10.48550/arXiv.2509.10789 arXiv:2509.10789 [cs]

    work page doi:10.48550/arxiv.2509.10789 2025

  59. [60]

    Lei Huang, Weijiang Yu, Weitao Ma, Weihong Zhong, Zhangyin Feng, Haotian Wang, Qianglong Chen, Weihua Peng, Xiaocheng Feng, Bing Qin, and Ting Liu. 2025. A Survey on Hallucination in Large Language Models: Principles, Taxonomy, Challenges, and Open Questions.ACM Transactions on Information Systems43, 2 (2025), 1–55. doi:10.1145/3703155

    work page doi:10.1145/3703155 2025

  60. [61]

    Tao Huang. 2025. Content moderation by LLM: from accuracy to legitimacy.Artificial Intelligence Review58, 10 (July 2025), 320. doi:10.1007/s10462-025-11328-1

    work page doi:10.1007/s10462-025-11328-1 2025

  61. [62]

    Victor Huang, Yifan Hu, and Yaohua Li. 2022. A Systematic Literature Review of New Trends in Self-expression Caused by Emojis and Memes:. InProceedings of the Conference. Atlantis Press, Sanya, China, 75–79. doi:10.2991/assehr.k.220105.016

    work page doi:10.2991/assehr.k.220105.016 2022

  62. [63]

    Jun Li Jeung and Janet Yi-Ching Huang. 2023. Correct Me If I Am Wrong: Exploring How AI Outputs Affect User Perception and Trust. InComputer Supported Cooperative Work and Social Computing(Minneapolis MN USA, 2023-10-14). ACM, New York, NY, USA, 323–327. doi:10.1145/3584931.3606997

    work page doi:10.1145/3584931.3606997 2023

  63. [64]

    Shagun Jhaver, Alice Qian Zhang, Quan Ze Chen, Nikhila Natarajan, Ruotong Wang, and Amy X. Zhang. 2023. Personalizing Content Moderation on Social Media: User Perspectives on Moderation Choices, Interface Design, and Labor.Proceedings of the ACM on Human-Computer Interaction7, CSCW2 (Sept. 2023), 1–33. doi:10.1145/3610080

    work page doi:10.1145/3610080 2023

  64. [65]

    Shagun Jhaver and Amy X. Zhang. 2025. Do users want platform moderation or individual control? Examining the role of third- person effects and free speech support in shaping moderation preferences.New Media & Society27, 5 (May 2025), 2930–2950. doi:10.1177/14614448231217993 Publisher: SAGE Publications

    work page doi:10.1177/14614448231217993 2025

  65. [66]

    Brubaker, and Casey Fiesler

    Jialun Aaron Jiang, Peipei Nie, Jed R. Brubaker, and Casey Fiesler. 2023. A Trade-off-centered Framework of Content Moderation.ACM Transactions on Computer-Human Interaction30, 1 (Feb. 2023), 1–34. doi:10.1145/3534929

    work page doi:10.1145/3534929 2023

  66. [67]

    Harmanpreet Kaur, Eytan Adar, Eric Gilbert, and Cliff Lampe. 2022. Sensible AI: Re-imagining interpretability and explainability using sensemaking theory. InProceedings of the 2022 ACM Conference on Fairness, Accountability, and Transparency. 702–714

    work page 2022

  67. [68]

    2022.Lawful but A wful? Control over Legal Speech by Platforms, Governments, and Internet Users

    Daphne Keller. 2022.Lawful but A wful? Control over Legal Speech by Platforms, Governments, and Internet Users. https://lawreviewblog. uchicago.edu/2022/06/28/keller-control-over-speech/

    work page 2022

  68. [69]

    John Kingston. 2017. Using Artificial Intelligence to Support Compliance with the General Data Protection Regulation.Artificial Intelligence and Law25, 4 (2017), 429–443. doi:10.1007/s10506-017-9206-9 Published online 1 September 2017

    work page doi:10.1007/s10506-017-9206-9 2017

  69. [70]

    Michał Klincewicz, Mark Alfano, and Amir Ebrahimi Fard. 2025. Slopaganda: The interaction between propaganda and generative AI. doi:10.48550/arXiv.2503.01560 arXiv:2503.01560 [cs]

    work page doi:10.48550/arxiv.2503.01560 2025

  70. [71]

    Rafal Kocielnik, Saleema Amershi, and Paul N. Bennett. 2019. Will You Accept an Imperfect AI?: Exploring Designs for Adjusting End-user Expectations of AI Systems. InProceedings of the 2019 CHI Conference on Human Factors in Computing Systems(2019-05-02). ACM, Glasgow, Scotland, UK, 1–14. doi:10.1145/3290605.3300641

    work page doi:10.1145/3290605.3300641 2019

  71. [72]

    Moritz Koerber. 2019. Theoretical considerations and development of a questionnaire to measure trust in automation. InProceedings of the 20th Congress of the International Ergonomics Association (IEA 2018): Volume VI: Transport Ergonomics and Human Factors (TEHF), Aerospace Human Factors and Ergonomics, S. Bagnara, R. Tartaglia, T. Albolino, T. Alexander,...

    work page 2019

  72. [73]

    Ivica Kostric, Krisztian Balog, and Ujwal Gadiraju. 2025. Should We Tailor the Talk? Understanding the Impact of Conversational Styles on Preference Elicitation in Conversational Recommender Systems. InProceedings of the 33rd ACM Conference on User Modeling, Adaptation Supporting Illegal Content Reporting under the Digital Services Act with LLMs FAccT ’26...

    work page doi:10.1145/3699682.3728353 2025

  73. [74]

    Yubo Kou and Xinning Gui. 2021. Flag and Flaggability in Automated Moderation: The Case of Reporting Toxic Behavior in an Online Game Community. InProceedings of the 2021 CHI Conference on Human Factors in Computing Systems (CHI ’21). Association for Computing Machinery, New York, NY, USA, 1–12. doi:10.1145/3411764.3445279

    work page doi:10.1145/3411764.3445279 2021

  74. [75]

    Deepak Kumar, Yousef AbuHashem, and Zakir Durumeric. 2024. Watch Your Language: Investigating Content Moderation with Large Language Models. doi:10.48550/arXiv.2309.14517 arXiv:2309.14517 [cs]

    work page doi:10.48550/arxiv.2309.14517 2024

  75. [76]

    Alisa Küper and Nicole Krämer. 2025. Psychological traits and appropriate reliance: Factors shaping trust in AI.International Journal of Human–Computer Interaction41, 7 (2025), 4115–4131

    work page 2025

  76. [77]

    Jinqi Lai, Wensheng Gan, Jiayang Wu, Zhenlian Qi, and Philip S. Yu. 2024. Large Language Models in Law: A Survey.AI Open5 (2024), 181–196. doi:10.1016/j.aiopen.2024.09.002

    work page doi:10.1016/j.aiopen.2024.09.002 2024

  77. [78]

    Vivian Lai, Samuel Carton, Rajat Bhatnagar, Q Vera Liao, Yunfeng Zhang, and Chenhao Tan. 2022. Human-ai collaboration via conditional delegation: A case study of content moderation. InProceedings of the 2022 CHI Conference on Human Factors in Computing Systems. 1–18

    work page 2022

  78. [79]

    Landgerichts Berlin. 2020. Beschluss vom 11.03.2020. https://openjur.de/u/2253932.html

    work page arXiv 2020

  79. [80]

    Florian Leiser, Sven Eckhardt, Valentin Leuthe, Merlin Knaeble, Alexander Mädche, Gerhard Schwabe, and Ali Sunyaev. 2024. HILL: A Hallucination Identifier for Large Language Models. InProceedings of the CHI Conference on Human Factors in Computing Systems (Honolulu HI USA, 2024-05-11). ACM, Honolulu, HI, USA, 1–13. doi:10.1145/3613904.3642428

    work page doi:10.1145/3613904.3642428 2024

  80. [81]

    Mark Leiser and Cristiana Santos. 2023. Dark Patterns, Enforcement, and the emerging Digital Design Acquis: Manipulation beneath the Interface.Dark Patterns, Enforcement, and the emerging Digital Design Acquis: Manipulation beneath the Interface(May 2023), 1–31. Preprint, not yet assigned to a journal volume or number. Publisher: SSRN

    work page 2023

Showing first 80 references.