# Detecting CSAM Text-to-Image LoRAs From Weights

arXiv: https://arxiv.org/abs/2607.25750 · HTML: https://mlsanity.com/paper/2607.25750/

## Key facts

- No citing papers in the map yet (data through 14 Oct 2026).
- Builds on 17 of its 40 references that are in the map.
- First posted on arXiv 28 Jul 2026.

## Builds on (most cited first, 17 of 17)

- [DreamBooth: Fine Tuning Text-to-Image Diffusion Models for Subject-Driven Generation](https://mlsanity.com/paper/2208.12242/) (2022)
- [Describing Textures in the Wild](https://mlsanity.com/paper/1311.3618/) (2013)
- [EuroSAT: A Novel Dataset and Deep Learning Benchmark for Land Use and Land Cover Classification](https://mlsanity.com/paper/1709.00029/) (2017)
- [Benchmarking Representation Learning for Natural World Image Collections](https://mlsanity.com/paper/2103.16483/) (2021)
- [TrojDiff: Trojan Attacks on Diffusion Models with Diverse Targets](https://mlsanity.com/paper/2303.05762/) (2023)
- [AI-Face: A Million-Scale Demographically Annotated AI-Generated Face Dataset and Fairness Benchmark](https://mlsanity.com/paper/2406.00783/) (2024)
- [Deepfakes on Demand: the rise of accessible non-consensual deepfake image generators](https://mlsanity.com/paper/2505.03859/) (2025)
- [Learning to Interpret Weight Differences in Language Models](https://mlsanity.com/paper/2510.05092/) (2025)
- [Seeing without Looking: Analysis Pipeline for Child Sexual Abuse Datasets](https://mlsanity.com/paper/2204.14110/) (2022)
- [A LoRA is Worth a Thousand Pictures](https://mlsanity.com/paper/2412.12048/) (2024)
- [Neglected Risks: The Disturbing Reality of Children's Images in Datasets and the Urgent Call for Accountability](https://mlsanity.com/paper/2504.14446/) (2025)
- [AI Generated Child Sexual Abuse Material -- What's the Harm?](https://mlsanity.com/paper/2510.02978/) (2025)
- [Unveiling AI's Threats to Child Protection: Regulatory efforts to Criminalize AI-Generated CSAM and Emerging Children's Rights Violations](https://mlsanity.com/paper/2503.00433/) (2025)
- [Stop the Nonconsensual Use of Nude Images in Research](https://mlsanity.com/paper/2510.22423/) (2025)
- [When LoRA Betrays: Backdooring Text-to-Image Models by Masquerading as Benign Adapters](https://mlsanity.com/paper/2602.21977/) (2026)
- ["Unlimited Realm of Exploration and Experimentation": Methods and Motivations of AI-Generated Sexual Content Creators](https://mlsanity.com/paper/2601.21028/) (2026)
- [Evaluation without Generation: Non-Generative Assessment of Harmful Model Specialization with Applications to CSAM](https://mlsanity.com/paper/2604.25119/) (2026)

## Abstract (arXiv)

Low-rank adaptation (LoRA) fine-tuning has made it cheap and easy to customize open-weight image generation models for specific tasks, including the production of child sexual abuse material (CSAM). Existing moderation relies on metadata or generated outputs, but metadata can be deceptive and generating outputs may itself be unacceptable or illegal. We show that a safer signal lives in the weights. The top-left singular vectors of a LoRA's updates form a compact, inference-free fingerprint (u1) of its strongest learned change. Using human-subject age as a benign proxy for CSAM, we find that u1 identifies what a LoRA was trained on, generalizes across base models, and abstains on unrelated benign content. The signal is robust to additive weight noise, rescaling, and precision reduction. These results indicate that harmful LoRAs could be screened directly from their weights without relying on metadata or generating harmful outputs.