EUVIMEDEUVIMED entdecken
Crossref

Deceptive Design Forensics — AI-Powered Detection of Privacy Interface Manipulation in Mobile Apps

Muhammad Junaid

Journal of Engineering, Science and Technological Trends · 20.11.2026 · 26-36

DOI: 10.64060/jestt3i34 ↗

Kurz gefasst

Fachartikel in der Zeitschrift Journal of Engineering, Science and Technological Trends, online erschienen am 20.11.2026. Verfasst von Muhammad Junaid.

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

Dark patterns are misleading User Interface (UI) designs that lead users to make decisions that are not in their best interests, especially in privacy areas that rely on the right to make choices. Although there are regulations like GDPR and CCPA, they still exist in mobile applications. The aim of this study is to see if a pre-trained CLIP vision encoder can be fine-tuned to classify screenshots of mobile app screens as a dark pattern or a clean interface. We were able to create a balanced set of 2,958 screenshots, 1,479 screenshots per class, that were derived from RICO screenshots with existing annotations and manually captured screenshots from around 35 apps. Only fine-tuning the last 6 layers of CLIP ViT-B/32 yielded a best accuracy of 79.90% on a 592-screenshot split of the data used for validation and to select the model, and 73.0% when the validation data split was used for an entire data split including training images. As the split was random at the screenshot level and there was no independent test set, neither of these is proof of generalisation to novel apps. The effect of class balancing, unfrozen layers and mild augmentation is shown through the Ablations. A locally-deployed inference service was assessed in a usability study involving 15 participants, with screenshots being sent to the inference service by a proof-of-concept Android client; execution crashed on the device, 3–5 per screenshot. The model is based on visual input only and can only identify deceptive intent, contextual manipulation or legal non-compliance in single screenshots.

Zitieren

APA

Junaid, M. (2026). Deceptive Design Forensics — AI-Powered Detection of Privacy Interface Manipulation in Mobile Apps. Journal of Engineering, Science and Technological Trends, 26-36. https://doi.org/10.64060/jestt3i34

Vancouver

Junaid M. Deceptive Design Forensics — AI-Powered Detection of Privacy Interface Manipulation in Mobile Apps. Journal of Engineering, Science and Technological Trends. 2026:26-36. doi:10.64060/jestt3i34

BibTeX
@article{Junaid2026,
  author = {Junaid, Muhammad},
  title = {Deceptive Design Forensics — AI-Powered Detection of Privacy Interface Manipulation in Mobile Apps},
  journal = {Journal of Engineering, Science and Technological Trends},
  year = {2026},
  pages = {26--36},
  doi = {10.64060/jestt3i34}
}
Kostenlos starten →