Pith. sign in

REVIEW 4 major objections 4 minor 49 references

50 Shades of Deceptive Patterns: A Unified Taxonomy, Multimodal Detection, and Security Implications

T0 review · 4 major / 4 minor · reviewed 2026-08-10 · deepseek-v4-flash

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

Pith's one-line read DPGuard, a binary classifier plus multimodal LLM with mutation-evolved prompts, outperforms prior deceptive-pattern tools and finds patterns in 23.61% of mobile and 47.27% of website screenshots.

desk verdict Useful dataset and taxonomy, but the SOTA claim rests on an unfair comparison and the wild prevalence numbers need manual validation. read the letter →

arxiv 2501.13351 v3 pith:HKER5RX2 submitted 2025-01-23 cs.CR

Zewei Shi , Ruoxi Sun , Jieshan Chen , Jiamou Sun , Minhui Xue , Yansong Gao , Feng Liu , Xingliang Yuan This is my paper · ORCID
classification cs.CR
keywords deceptivepatternsdarktaxonomyDPGuardmultimodallargelanguagemodelUIscreenshotdetectionprivacyempiricalstudy
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

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

The reading

This paper tries to establish that deceptive patterns in mobile apps and websites can be mapped onto one unified taxonomy and detected automatically by a two-stage tool called DPGuard. The tool first uses a fine-tuned image classifier to decide whether a screenshot is deceptive, then hands only suspicious screenshots to a multimodal large language model whose classification prompt has been improved through mutation. Evaluated on a newly assembled dataset of 6,725 images containing 10,421 labeled pattern instances, DPGuard reports higher micro and macro F1 scores than the two previous state-of-the-art detectors, UIGuard and AidUI. Applying DPGuard to 2,000 popular apps and websites, the authors report that 23.61% of mobile screenshots and 47.27% of website screenshots contain at least one deceptive pattern instance. If these numbers hold, regulators and platform reviewers would have a practical way to audit interface manipulation at scale.

What carries the argument

The central mechanism is DPGuard's two-stage inference pipeline: a fine-tuned ResNet101 binary classifier decides whether a screenshot is deceptive, and only positively classified images are passed to GPT-4o, a multimodal LLM that names the specific deceptive-pattern category. The prompt used by the LLM is produced by prompt mutation, an adaptation of PromptBreeder in which the model paraphrases, adds, or deletes actions; a quality checker keeps mutated prompts whose cosine similarity to the initial prompt exceeds a threshold, and a prompt queue retains the best-performing prompts across mutation rounds using binary cross-entropy loss on a balanced batch. The unified 21-category taxonomy is the annotation scheme that ties the dataset, the system prompt, and the evaluation together, defining what counts as a deceptive pattern instance.

What would settle it

Recompute Table 6 using only the categories that UIGuard and AidUI support, so all three tools are scored on the identical subset of instances; if DPGuard's micro and macro F1 no longer exceed the baselines on that common subset, the 'outperforms state-of-the-art' claim reduces to a taxonomy-coverage artifact and the detection advantage disappears.

Watch

Extended reading notes

Core claim

The paper's central claim is that a two-stage pipeline—a low-cost binary classifier that filters out non-deceptive screenshots, followed by a commercial multimodal large language model guided by a mutation-evolved prompt—is sufficient to detect deceptive patterns at the state of the art. On the authors' cross-platform dataset, DPGuard achieves micro/macro F1 of 0.73/0.44 on mobile screenshots and 0.50/0.34 on website screenshots, exceeding UIGuard and AidUI on both. The paper also claims that its refined 21-category taxonomy, which reintroduces Forced Enrollment and expands the scope of five existing categories, is necessary because previous taxonomies missed security- and privacy-relevant designs such as undisclosed subscription fees, fake scarcity, and privacy terms buried in hyperlinks. Under this taxonomy, the empirical study of 2,000 popular services finds that 23.61% of mobile screenshots and 47.27% of website screenshots contain at least one deceptive pattern, with websites averaging more instances per screen than mobile apps.

Load-bearing premise

The comparison assumes that scoring UIGuard and AidUI on a dataset annotated with a different taxonomy—with many categories marked unsupported for those tools—still yields micro and macro F1 averages comparable to DPGuard's, so the reported state-of-the-art advantage could be an artifact of taxonomy coverage rather than detection skill.

Editorial extensions

If this is right

  • Because DPGuard only queries the LLM for screenshots the binary classifier flags, bulk auditing of app stores or the web becomes far cheaper than running an LLM on every image.
  • Since the taxonomy lives in the system prompt, updating the detector to a revised or extended taxonomy only requires re-running the prompt-mutation loop, which is the paper's intended answer to concept drift in deceptive designs.
  • The reported prevalence numbers imply that deceptive patterns are not rare edge cases: nearly half of website screenshots and roughly a quarter of mobile screenshots from popular services carry at least one instance, so any platform-level mitigation would affect a large fraction of user interactions.
  • The four scope expansions give security practitioners concrete new pattern types to test for, such as post-trial subscription fees, fake scarcity countdowns, privacy terms hidden in hyperlinks, and plan-comparison barriers.
  • Removing Bait-and-Switch while adding Forced Enrollment changes what a detector will flag; suites of screenshots from prior studies may need re-annotation before results from different taxonomies can be compared.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • The same two-stage design could be repurposed for other visual-manipulation audits, such as cookie-consent banners, in-app purchase flows, or political advertising, by swapping in a domain-specific taxonomy in the system prompt.
  • The website prevalence figure of 47.27% is probably an underestimate: the crawler visits a limited number of pages per domain and cannot trigger dynamic consent modals, checkout stages, or post-trial states where many hidden costs actually appear.
  • The reported F1 numbers are tied to one commercial model snapshot; if future GPT-4o updates change behavior, the exact scores will drift, so the durability of the framework should be assessed by re-running the mutation loop on the same data rather than by fixing confidence to the current metric values.
  • An open-weights model fine-tuned on the same dataset could test how much of the gain comes from the prompt-mutation method versus the underlying commercial model, and would also make the detector usable on interfaces where screenshots cannot be sent to external APIs.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

4 major / 4 minor

Summary. This paper presents a unified taxonomy of deceptive patterns with 21 categories, a new dataset of 6,725 UI images containing 10,421 DP instances, and DPGuard, a hybrid detection framework that combines a fine-tuned binary classifier (ResNet101) with a commercial multimodal LLM (GPT-4o) whose prompts are refined through a mutation-based prompt engineering process. The authors report that DPGuard outperforms existing SOTA detectors UIGuard and AidUI on both mobile and website datasets, and they present a large-scale empirical study of 12,301 wild UI images claiming that 23.61% of mobile screenshots and 47.27% of website screenshots contain at least one deceptive pattern. The paper also includes four case studies linking the expanded taxonomy categories to security and privacy implications.

Significance. If the reported results are valid, the paper would provide a substantial community resource: a large cross-platform deceptive-pattern dataset, a unified taxonomy that explicitly incorporates privacy and security concerns, and a working automatic detector that could scale beyond the small manually curated corpora used by prior work. The prompt-mutation approach to zero-shot MLLM adaptation is an interesting engineering contribution, and the code is released. The wild prevalence figures, if independently validated, would be an important quantification of deceptive design in popular apps and websites. However, the significance is contingent on resolving several evaluation and validation concerns, most notably the construction of the SOTA comparison in Table 6 and the lack of independent ground-truth validation for the in-the-wild measurements.

major comments (4)
  1. [Section 5.1, Table 6] The overall DPGuard evaluation in Table 6 does not specify which data split is used. The 'Instances' column sums to the full dataset counts (after removing the Sneak into Basket and Tricked Questions categories), rather than to a held-out test set. Since the binary classifier was fine-tuned on a 6:2:2 split and the prompt mutation process selected prompts using batches from the training portion (Appendix A), an evaluation on the full dataset would allow training and prompt-selection data to be present in the evaluation set, potentially inflating DPGuard's reported F1 scores. The paper must state the exact evaluation split and report results on the held-out test set (e.g., the 20% reserved for the MLLM evaluation) separately from any in-distribution numbers.
  2. [Table 6, Takeaway 2] The macro F1 averages for UIGuard and AidUI are computed over the full 22-class taxonomy with unsupported categories ('-') treated as zero. This is verifiable: UIGuard's mobile macro F1 of 0.2851 equals 6.2712/22 (the sum of its 13 supported F1 values divided by 22), and AidUI's 0.0878 equals 1.9323/22; the website rows follow the same pattern. Consequently, the reported SOTA advantage conflates taxonomy coverage with detection capability. The comparison should be restricted to the common support set, or the baselines should be adapted/retrained to the unified taxonomy. A restricted recomputation over the UIGuard-supported mobile classes reduces the macro gap from about 0.153 to about 0.060, so the specific Takeaway 2 numbers are not established as like-for-like SOTA improvements.
  3. [Section 5.2 and Appendix B.1] The wild prevalence figures (23.61% of mobile screenshots and 47.27% of website screenshots) are DPGuard's own predictions on the collected wild data. Appendix B.1 states that the authors 'randomly sampled some data for manual review to assess the actual performance of our model in the wild,' but no such validation results are reported anywhere in the paper. Without independent ground-truth labels on a random sample of the wild images, these prevalence rates are model outputs, not measurements. The paper should report the manual review outcome (e.g., precision and recall on the wild sample) and present calibrated prevalence estimates, or clearly label the figures as model-predicted rates.
  4. [Section 3 (Dataset creation) and Appendix A] The taxonomy definitions, the dataset labels, and the evaluation labels are all produced by the same authors, and no annotation protocol or inter-annotator agreement is reported. Because DPGuard is optimized and evaluated on these self-produced labels, the reported F1 scores may partly reflect the authors' interpretation of the taxonomy rather than an objective ground truth. The paper should provide annotation guidelines, inter-annotator agreement statistics, and ideally an external validation set (e.g., from prior taxonomies or independent reviewers) to support both the SOTA performance claim and the prevalence measurements.
minor comments (4)
  1. [Abstract (front matter vs. paper text)] The front-matter abstract reports prevalence as '23.61% of mobile screenshots and 47.27% of website screenshots,' while the paper's abstract section reports '25.7% of mobile apps and 49.0% of websites.' These are different metrics (image-level vs. app/domain-level), and the headline claim should use one consistent quantity or explicitly present both.
  2. [Section 7] The conclusion states that the taxonomy was 'refining it with 24 subcategories,' but Table 2 lists 21 DP categories. Please reconcile this number.
  3. [Algorithm 1 and Section 4.3] The text says prompt mutation terminates if the best prompt has not been updated for three rounds, but the pseudocode in Algorithm 1 loops until t<T without a break for that condition. Make the pseudocode consistent with the stated termination rule.
  4. [Section 6.4] The text contains a typo: 'Fiugre 9(c)' should be 'Figure 9(c)'.

Circularity Check

1 steps flagged · score 6.0 of 10

Table 6's state-of-the-art comparison is partly constructed: unsupported baseline categories are scored as zero in the macro F1, so DPGuard's claimed advantage over UIGuard and AidUI conflates taxonomy coverage with detection skill.

  1. other [Section 5.1, Table 6 and Takeaway 2]
    "-: the DP category is not supported by the corresponding tool. ... Macro avg 7,114 0.2851 0.0878 0.4385 ... Takeaway 2: DPGuard outperforms the state-of-the-art models in DP detection, increasing the F1-score to 0.73 (micro) and 0.44 (macro) on the mobile dataset, and 0.50 (micro) and 0.34 (macro) on the website dataset."

    The reported macro F1 for UIGuard and AidUI is computed over the paper's own 22-class taxonomy, including categories marked '-' as unsupported. Concretely, UIGuard's mobile macro F1 of 0.2851 equals the sum of its 13 non-'-' category F1 values (6.2712) divided by 22, and AidUI's 0.0878 equals 1.9323 divided by 22; the website rows follow the same pattern. Unsupported categories are thus scored as F1=0 even though those tools were never designed to express them (e.g., Forced Enrollment, Intermediate Currency, Hidden Costs). The claimed SOTA advantage is therefore partly an artifact of the evaluation construction: DPGuard is penalized for no category it cannot express, while each baseline is penalized for every instance in categories outside its taxonomy.

full rationale

The core taxonomy construction, dataset annotation, and DPGuard training/evaluation are largely self-contained: the prompt mutation uses a reserved test split, the binary classifier is tuned on held-out validation data, and the authors' own labeled dataset is a legitimate test bed. The self-citations to UIGuard are not load-bearing in a proof sense because the paper explicitly modifies the prior taxonomy rather than invoking it as an external uniqueness result. The most defensible circularity is in the SOTA comparison: the macro-F1 arithmetic in Table 6 shows that unsupported baseline categories are counted as zero, making part of DPGuard's reported performance advantage an artifact of the authors' taxonomy and scoring convention. The wild prevalence rates in Section 5.2 are also presented as measurements while the stated manual-review validation is not reported, which is a serious limitation, but it is an unvalidated prediction rather than a formal circular reduction.

Assumptions & free parameters 8 free parameters · 6 assumptions · 0 invented entities

The central claims rest on a substantial set of fitted thresholds, the authors' own annotations, and the commercial MLLM's behavior. The free parameters are all tuned on small pilots or chosen for cost reasons, none are derived from first principles. The key untested assumptions are taxonomy validity, annotation correctness, and the fairness of the baseline comparison.

free parameters (8)
  • Prompt similarity threshold s = 0.2
    Set via a pilot study of 90 prompts (30 mutated prompts over 3 rounds) manually grouped into good and poor quality. Used in Algorithm 1 line 7 to filter mutated prompts.
  • Prompt queue size n = 15
    Chosen to bound MLLM query cost while retaining diverse prompts across mutation rounds.
  • Total mutation rounds T = 25
    Stopping criterion with early termination if no update for 3 rounds; selected through pilot experiments.
  • Batch size b for prompt evaluation = 100
    Balanced batch with at least 5 examples per category to limit API cost.
  • Image dedup threshold step 2 = 0.95
    Tuned on 20 manually reviewed APKs to retain 172 of 175 expected unique images.
  • Image dedup threshold step 3 = 0.90
    Tuned on the same 20 APKs to retain 116 of 137 expected meaningful images.
  • Website screenshot file size filter = 8 KB
    Threshold chosen by inspection to drop blank or near-empty pages.
  • ResNet101 fine-tuned weights = Trained to F1 0.8769 on binary DP task
    The binary classifier is the first stage of DPGuard; its parameter values are fitted on the new dataset.
assumptions (6)
  • domain assumption The 21-category taxonomy is complete and categories are disjoint.
    All labeling, evaluation, and prevalence claims depend on this taxonomy being a faithful operationalization of deceptive patterns; no independent validation is provided.
  • domain assumption The authors' manual annotations of the merged dataset are correct.
    The dataset ground truth is the authors' labeling, with no inter-annotator agreement study or second-labeler check reported.
  • domain assumption GPT-4o can reliably interpret UI screenshots for deceptive patterns.
    The MLLM is the categorical engine of DPGuard; its standalone micro F1 is only 0.50 to 0.53 (Table 5), so the detector's final performance relies on the binary classifier filtering and prompt engineering.
  • domain assumption The random 6:2:2 split does not leak near-duplicate images from the same app or same source dataset across train and test.
    The dataset merges images from UIGuard, AidUI, and LLE, and new crawls from the same apps could appear in both splits; the paper does not deduplicate across splits.
  • domain assumption The in-the-wild screenshots are representative of popular mobile apps and websites.
    AndroZoo apps were randomly selected without ranking, and Majestic Million top 1,000 is a ranking list, so the mobile sample may not be 'popular' in the same sense as the web sample.
  • ad hoc to paper The similarity threshold from the pilot study (s=0.2) generalizes to the full mutation process.
    The threshold was derived from 90 prompts in a small pilot and then fixed for all experiments.

how reviews work

0 comments
Cite this review

Pith. "Pith review of 50 Shades of Deceptive Patterns: A Unified Taxonomy, Multimodal Detection, and Security Implications." pith.science (2026). https://pith.science/paper/HKER5RX2

@misc{pith2026250113351,
  author       = {Pith},
  title        = {Pith review of: 50 Shades of Deceptive Patterns: A Unified Taxonomy, Multimodal Detection, and Security Implications},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/HKER5RX2}},
  note         = {Machine review of arXiv:2501.13351}
}
read the original abstract

Deceptive patterns (DPs) are user interface designs deliberately crafted to manipulate users into unintended decisions, often by exploiting cognitive biases for the benefit of companies or services. While numerous studies have explored ways to identify these deceptive patterns, many existing solutions require significant human intervention and struggle to keep pace with the evolving nature of deceptive designs. To address these challenges, we expanded the deceptive pattern taxonomy from security and privacy perspectives, refining its categories and scope. We created a comprehensive dataset of deceptive patterns by integrating existing small-scale datasets with new samples, resulting in 6,725 images and 10,421 DP instances from mobile apps and websites. We then developed DPGuard, a novel automatic tool leveraging commercial multimodal large language models (MLLMs) for deceptive pattern detection. Experimental results show that DPGuard outperforms state-of-the-art methods. Finally, we conducted an extensive empirical evaluation on 2,000 popular mobile apps and websites, revealing that 23.61% of mobile screenshots and 47.27% of website screenshots feature at least one deceptive pattern instance. Through four unexplored case studies that inform security implications, we highlight the critical importance of the unified taxonomy in addressing the growing challenges of Internet deception.

Figures

Figures reproduced from arXiv: 2501.13351 by the authors.

Figure 1
Figure 1. An overview of DPGuard framework. Algorithm 1: A Mutation-based Prompt Engineering Input: Initial prompt 𝑃0, System prompt 𝑃𝑠 , Mutation instructions 𝑃𝑚, Prompt queue 𝑄, Queue size limit 𝑛, Number of new prompts to be generated in each round 𝑚, Training dataset 𝐷, Total mutation rounds 𝑇 , Multimodal language model 𝑀, Batch size 𝑏, Similarity threshold 𝑠. Output: Best prompt 𝑝𝑏 . 1 𝑝𝑏 ← 𝑃0, 𝑄 ← ∅, 𝑡 ← 0; 2 𝑄.enqueue… view at source ↗
Figure 2
Figure 2. Loss in each mutation round. DP category determining through MLLM. In our prompt muta￾tion strategy, there are three hyper-parameters: similarity threshold, queue limits, and the number of mutation rounds. We select these hyper-parameters through several pilot experiments. To demon￾strate the performance of our prompt engineering, we use GPT4o with a fixed initial prompt as a baseline, and compare it to the final be… view at source ↗
Figure 3
Figure 3. Distribution of number of deceptive instances per [PITH_FULL_IMAGE:figures/full_fig_p007_3.png] view at source ↗
Figures from the paper (6 more)
Figure 4
Figure 4. Figure 4: Distribution of detected deceptive instances in the [PITH_FULL_IMAGE:figures/full_fig_p007_4.png]
Figure 5
Figure 5. Figure 5: Reasons of removing screenshots in step3 [PITH_FULL_IMAGE:figures/full_fig_p011_5.png]
Figure 6
Figure 6. Figure 6: Number of images distributed in an app [PITH_FULL_IMAGE:figures/full_fig_p011_6.png]
Figure 7
Figure 7. Figure 7: Status code and error count [PITH_FULL_IMAGE:figures/full_fig_p011_7.png]
Figure 8
Figure 8. Figure 8: The number of visited webpages within a domain. [PITH_FULL_IMAGE:figures/full_fig_p011_8.png]
Figure 9
Figure 9. Figure 9: Examples of deceptive patterns [PITH_FULL_IMAGE:figures/full_fig_p012_9.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

49 extracted references · 39 canonical work pages

  1. [1]

    Bissyandé, Jacques Klein, and Yves Le Traon

    Kevin Allix, Tegawendé F. Bissyandé, Jacques Klein, and Yves Le Traon. 2016. AndroZoo: Collecting millions of Android apps for the research community. In Proceedings of the 13th International Conference on Mining Software Repositories . 468–471

    work page 2016

  2. [2]

    Anthropic. 2024. Claude. https://claude.ai/

    work page 2024

  3. [3]

    Nataliia Bielova, Laura Litvine, Anysia Nguyen, Mariam Chammat, Vincent Toubiana, and Estelle Hary. 2024. The effect of design patterns on (present and future) cookie consent decisions. In 33rd USENIX Security Symposium (USENIX Security 24). 2813–2830

    work page 2024

  4. [4]

    European Data Protection Board. 2022. EDPB adopts guidelines on Art. 60 GDPR, guidelines on dark patterns in social media platform interfaces, toolbox on essential data protection safeguards for enforcement cooperation between EEA and third country SAs. https://edpb.europa.eu/news/news/2022/edpb-adopts- guidelines-art-60-gdpr-guidelines-dark-patterns-soci...

    work page 2022

  5. [5]

    I am definitely manipulated, even when I am aware of it. It’s ridiculous!

    Kerstin Bongard-Blanchy, Arianna Rossi, Salvador Rivas, Sophie Doublet, Vincent Koenig, and Gabriele Lenzini. 2021. “I am definitely manipulated, even when I am aware of it. It’s ridiculous!” - Dark patterns from the end-user perspective. In Proceedings of the 2021 ACM Designing Interactive Systems Conference . 763–776

    work page 2021

  6. [6]

    Harry Brignull. 2010. Twitter: @darkpatterns. https://x.com/darkpatterns

    work page 2010

  7. [7]

    Harry Brignull, Mark Leiser, Cristiana Santos, and Kosha Doshi. 2010. Deceptive patterns: User interfaces designed to trick people. https://old.deceptive.design /main_page/index.html

    work page 2010

  8. [8]

    Jieshan Chen, Jiamou Sun, Sidong Feng, Zhenchang Xing, Qinghua Lu, Xiwei Xu, and Chunyang Chen. 2023. Unveiling the tricks: Automated detection of dark patterns in mobile applications. In Proceedings of the 36th Annual ACM Symposium on User Interface Software and Technology . 1–20

    work page 2023

Show all 49 references
  1. [9]

    Wenliang Dai, Junnan Li, Dongxu Li, Anthony Tiong, Junqi Zhao, Weisheng Wang, Boyang Li, Pascale Fung, and Steven Hoi. 2023. InstructBLIP: Towards General-purpose Vision-Language Models with Instruction Tuning. In Proceed- ings of the 37th International Conference on Neural In...

    2023

  2. [10]

    Google DeepMind. 2024. Gemini Pro: Our best model for general performance across a wide range of tasks. https://deepmind.google/technologies/gemini/pro/

    2024

  3. [11]

    Linda Di Geronimo, Larissa Braz, Enrico Fregnan, Fabio Palomba, and Alberto Bacchelli. 2020. UI dark patterns and where to find them: A study on mobile applications and user perception. In Proceedings of the 2020 CHI Conference on Human Factors in Computing Systems . 1–14

    2020

  4. [12]

    California Privacy Protection Agency Enforcement Division. 2024. Avoiding dark patterns: clear and understandable language, symmetry in choice. https: //cppa.ca.gov/pdf/enfadvisory202402.pdf

    2024

  5. [13]

    Chrisantha Fernando, Dylan Banarse, Henryk Michalewski, Simon Osindero, and Tim Rocktäschel. 2023. Promptbreeder: Self-referential self-improvement via prompt evolution. arXiv preprint arXiv:2309.16797 (2023)

    2023 arXiv

  6. [14]

    Colin M Gray, Yubo Kou, Bryan Battles, Joseph Hoggatt, and Austin L Toombs

  7. [15]

    Kaiming He, Xiangyu Zhang, Shaoqing Ren, and Jian Sun. 2016. Deep residual learning for image recognition. InProceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR) . 770–778

    2016

  8. [16]

    Shun Hidaka, Sota Kobuki, Mizuki Watanabe, and Katie Seaborn. 2023. Linguistic dead-ends and alphabet soup: Finding dark patterns in japanese apps. In Pro- ceedings of the 2023 CHI Conference on Human Factors in Computing Systems . 1–13

    2023

  9. [17]

    Gao Huang, Zhuang Liu, Laurens Van Der Maaten, and Kilian Q Weinberger

  10. [18]

    Dixon Jones. 2012. Majestic million CSV now free for all, daily. Technical Report. Majestic

    2012

  11. [19]

    Gunawan, David Choffnes, Daniel J Dubois, Woodrow Hartzog, and Christo Wilson

    Monica Kowalczyk, Johanna T. Gunawan, David Choffnes, Daniel J Dubois, Woodrow Hartzog, and Christo Wilson. 2023. Understanding dark patterns in home IoT devices. In Proceedings of the 2023 CHI Conference on Human Factors in Computing Systems. 1–27

    2023

  12. [20]

    Jian Li and Weiheng Lu. 2024. A survey on benchmarks of multimodal large language models. arXiv preprint arXiv:2408.08632 (2024)

    2024 arXiv

  13. [21]

    Yuanchun Li, Ziyue Yang, Yao Guo, and Xiangqun Chen. 2017. DroidBot: A lightweight UI-guided test input generator for Android. In Proceedings of the 39th International Conference on Software Engineering Companion (ICSE-C) . 23–26

    2017

  14. [22]

    Haotian Liu, Chunyuan Li, Yuheng Li, Bo Li, Yuanhan Zhang, Sheng Shen, and Yong Jae Lee. 2024. LLaVA-NeXT: Improved reasoning, OCR, and world knowl- edge. https://llava-vl.github.io/blog/2024-01-30-llava-next/

    2024

  15. [23]

    S M Hasan Mansur, Sabiha Salma, Damilola Awofisayo, and Kevin Moran. 2023. AidUI: Toward automated recognition of dark patterns in user interfaces. In Proceedings of the 45th International Conference on Software Engineering . 1958– 1970

    2023

  16. [24]

    Arunesh Mathur, Gunes Acar, Michael J Friedman, Eli Lucherini, Jonathan Mayer, Marshini Chetty, and Arvind Narayanan. 2019. Dark patterns at scale: Findings from a crawl of 11K shopping websites. Proceedings of the ACM on Human- Computer Interaction 3, CSCW (2019), 1–32

    2019

  17. [25]

    Dmitry Nazarov and Yerkebulan Baimukhambetov. 2022. Clustering of dark pat- terns in the user interfaces of websites and online trading portals (E-Commerce). Mathematics 10, 18 (2022), 3219

    2022

  18. [26]

    Trung Tin Nguyen, Michael Backes, and Ben Stock. 2022. Freely given consent? Studying consent notice of third-Party tracking and its violations of GDPR in Android apps. In Proceedings of the 2022 ACM SIGSAC Conference on Computer and Communications Security. 2369–2383

    2022

  19. [27]

    Liming Nie, Yangyang Zhao, Chenglin Li, Xuqiong Luo, and Yang Liu. 2024. Shadows in the interface: A comprehensive study on dark patterns. Proceedings of the ACM on Software Engineering 1, FSE (2024), 204–225

    2024

  20. [28]

    Ikechukwu Obi, Colin M Gray, Shruthi Sai Chivukula, Ja-Nae Duane, Janna Johns, Matthew Will, Ziqing Li, and Thomas Carlock. 2022. Let’s talk about socio- technical angst: Tracing the history and evolution of dark patterns on twitter from 2010-2021. arXiv preprint arXiv:2207.10...

    2022 arXiv

  21. [29]

    Australian Federal Register of Legislation. 2024. Competition and consumer act

    2024

  22. [30]

    OpenAI. 2024. GPT-4o mini: Advancing cost-efficient intelligence. https://open ai.com/index/gpt-4o-mini-advancing-cost-efficient-intelligence/

    2024

  23. [31]

    OpenAI. 2024. Hello GPT-4o. https://openai.com/index/hello-gpt-4o/

    2024

  24. [32]

    PyTorch. 2024. BCELoss. https://pytorch.org/docs/stable/generated/torch.nn.B CELoss.html Accessed: 2024-10-13

    2024

  25. [33]

    Nils Reimers and Iryna Gurevych. 2019. Sentence-BERT: Sentence embeddings using siamese BERT-networks. In Proceedings of the 2019 Conference on Empirical Methods in Natural Language Processing and the 9th International Joint Conference on Natural Language Processing (EMNLP-IJCNLP)

    2019

  26. [34]

    Zewei Shi, Ruoxi Sun, Jieshan Chen, Jiamou Sun, and Minhui Xue. 2024. The invisible game on the Internet: A case study of decoding deceptive patterns. In Companion Proceedings of the ACM Web Conference 2024 . 521–524

    2024

  27. [35]

    Karen Simonyan and Andrew Zisserman. 2014. Very deep convolutional networks for large-scale image recognition. arXiv preprint arXiv:1409.1556 (2014)

    2014 arXiv

  28. [36]

    Ruoxi Sun, Jiamin Chang, Hammond Pearce, Chaowei Xiao, Bo Li, Qi Wu, Surya Nepal, and Minhui Xue. 2024. SoK: Unifying cybersecurity and cybersafety of multimodal foundation models with an information theory approach. arXiv preprint arXiv:2411.11195 (2024)

    2024

  29. [37]

    Dewen Suo, Lei Xue, Runze Tan, Weihao Huang, and Guozi Sun. 2024. ARAP: Demystifying anti runtime analysis code in Android apps. arXiv preprint arXiv:2408.11080 (2024)

    2024

  30. [38]

    Christine Utz, Martin Degeling, Sascha Fahl, Florian Schaub, and Thorsten Holz

  31. [39]

    Jason Wu, Siyan Wang, Siman Shen, Yi-Hao Peng, Jeffrey Nichols, and Jeffrey P Bigham. 2023. WebUI: A dataset for enhancing visual UI understanding with web semantics. In Proceedings of the 2023 CHI Conference on Human Factors in Computing Systems. 1–14

    2023

  32. [40]

    Qinge Xie and Frank Li. 2024. Crawling to the top: An empirical evaluation of top list use. In 25th International Conference on Passive and Active Network Measurement. 277–306

    2024

  33. [41]

    Chang Yue, Chen Zhong, Kai Chen, Zhiyu Zhang, and Yeonjoon Lee. 2024. DARK- FLEECE: Probing the dark Side of Android subscription apps. In 33rd USENIX Security Symposium (USENIX Security 24) . 1543–1560

    2024

  34. [42]

    Xiangyu Zhang, Lingling Fan, Sen Chen, Yucheng Su, and Boyuan Li. 2023. Scene-driven exploration and GUI modeling for Android apps. In Proceedings of the 38th IEEE/ACM International Conference on Automated Software Engineering . 1251–1262

    2023

  35. [43]

    Deyao Zhu, Jun Chen, Xiaoqian Shen, Xiang Li, and Mohamed Elhoseiny. 2024. MiniGPT-4: Enhancing Vision-Language Understanding with Advanced Large Language Models. In The Twelfth International Conference on Learning Represen- tations. Appendix A Dataset for DPGuard Evaluation I...

    2024

  36. [48]

    **Nagging**: Spot any repetitive and unexpected pop-up win- dows that disrupt user activities. 2. **Roach Motel**: Identify sce- narios where opting in is easy, but opting out is complicated or obscure. 3. **Price Comparison Prevention**: Detect any elements that hinder direct...

  37. [49]

    **Social Pyramid**: Look for incentives encouraging users to share content with friends for rewards. 16. **Privacy Zuckering**: Assess if default options necessitate sharing unnecessary personal information. 17. **Gamification**: Note if users are required to re- peatedly perf...

    2025

  38. [2010]

    https://www.legislation.gov.au/C2004A00109/latest/text

  39. [2017]

    In Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR)

    Densely connected convolutional networks. In Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR) . 4700–4708

  40. [2018]

    In Proceedings of the 2018 CHI Conference on Human Factors in Computing Systems

    The dark (patterns) side of UX design. In Proceedings of the 2018 CHI Conference on Human Factors in Computing Systems . 1–14

    2018

  41. [2019]

    In Proceedings of the 2019 ACM SIGSAC Conference on Computer and Communications Security

    (Un)informed consent: Studying GDPR consent notices in the field. In Proceedings of the 2019 ACM SIGSAC Conference on Computer and Communications Security. 973–990

    2019

Pith tools

Reviewed August 10, 2026 · model on record in the stance chip above.