Research map / Image generation / Autoregressive and latent generation
Stop the Nonconsensual Use of Nude Images in Research
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's full text. Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
A position paper that systematically reviews 150 computer-science papers using real nude images and argues the ML community must stop nonconsensually collecting and distributing such images in research.
Problem
Nudity detection research routinely scrapes, annotates, publishes and releases real nude images without subject consent, which perpetuates image-based sexual abuse and exposes subjects to further harm.
Approach
Applies consent, harm, and ethics frameworks (consequentialist and deontological) to a systematic annotation of 150 papers that use nude image datasets, documenting collection, publication, and distribution practices and proposing a call to action for venues and a consent-based data trust.
Key results
- Of 150 reviewed papers, 87 published example nude images in their papers, with nine leaving them uncensored and 28 censoring only bodies while leaving faces visible.
- Of the 98 papers that created new datasets, 60 did not name a specific data source and 19 did not state one; none of the 74 fully annotated papers discussed subject consent.
- Only two papers mentioned IRB approval and only eight discussed data security, while 51 of the framing-annotated papers cited reducing pornography as their aim.
Surprising findings
- Some papers reportedly collected content from sources like subreddits later banned for sexual violence, and one paper described collecting CSAM-like keyword searches, indicating researchers may be collecting abusive content.
- Researchers' stated justification for distributing nude images is often unsupported, and three papers that made data public still left nude datasets accessible on HuggingFace and figshare at submission.
Limitations
- Corpus limited to Google Scholar keyword search and DBLP-indexed venues, so it under-covers the full literature.
- Under-reporting of data sources means the extent of abusive content is likely underestimated.
- Annotations rely on what papers state, with some inferences about data origin drawn from descriptions alone.
Experimental setup
Datasets: NudeNet
Metrics: descriptive statistics of papers, datasets, and published im
Builds on: Kohno et al. ethics framework for computer security research
Closest papers by meaning
- Six-CD: Benchmarking Concept Removals for Text-to-image Diffusion Models (CVPR 2025 · similarity 0.82)
- Comprehensive Assessment and Analysis for NSFW Content Erasure in Text-to-Image Diffusion models (NeurIPS 2025 · similarity 0.81)
- Image Safeguarding: Reasoning with Conditional Vision Language Model and Obfuscating Unsafe Content Counterfactually (AAAI 2024 · similarity 0.81)
- PerceptAnon: Exploring the Human Perception of Image Anonymization Beyond Pseudonymization for GDPR (ICML 2024 · similarity 0.81)
- Mitigating Sexual Content Generation via Embedding Distortion in Text-conditioned Diffusion Models (NeurIPS 2025 · similarity 0.80)
- Enhancing Privacy in the Early Detection of Sexual Predators Through Federated Learning and Differential Privacy (AAAI 2025 · similarity 0.80)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.