Generative ML and CSAM: Implications and Mitigations
Open accessAbstract
A joint report between the Stanford Internet Observatory and Thorn examines implications of fully realistic child sexual abuse material (CSAM) produced by generative machine learning models. Advances in the open-source generative ML community have led to increasingly realistic adult content, to the point that content indistinguishable from actual photographs is likely to be common in the very near future. These same models and techniques have been also leveraged to produce CSAM. We examine what has enabled this state of affairs, the potential societal consequences of the proliferation of such content, and measures that can be taken to minimize harm from current and future visual generative ML models.
Integrity signals
Registry
Cites
Statistics
Claims the paper makes
Claims other papers make about this one
Each is checked against this paper’s own text: does it say what the citing paper says it does?
In the CSAM study (Thiel et al., 2023), photorealistic computer-generated CSAM makes up under 1% of CSAM found on online forums.
“they find that less than 1% of CSAM on online forums is photorealistic CG-CSAM.”
From On the Societal Impact of Open Foundation Models (2024) · cites this paper as Thiel et al., 2023
Watermarking or monitoring Stable Diffusion use is difficult because its weights are open.
“points out that watermarking or monitoring the use of Stable Diffusion is hard because of open weights”
From On the Societal Impact of Open Foundation Models (2024) · cites this paper as Thiel et al., 2023
The CSAM study finds that the bulk of CG-CSAM derives from Stable Diffusion models, which are open.
“The paper also provides evidence that most of CG-CSAM is based on the Stable Diffusion series of models, which are openly available”
From On the Societal Impact of Open Foundation Models (2024) · cites this paper as Thiel et al., 2023
Cited by
Sorted by citations
Parti, K., & Szabó, J. (2024) · Laws · doi:10.3390/laws13060067 · Cited by 15 · see citing works
On the Societal Impact of Open Foundation Models
Kapoor, S., Bommasani, R., Klyman, K., Longpre, S., Ramaswami, A., Cihon, P., Hopkins, A., Bankston, K., Biderman, S., Bogen, M., et al. (2024) · arXiv · doi:10.48550/arxiv.2403.07918 · Cited by 11 · see citing works
- Authors
- Thiel, D., Stroebel, M., Portnoff, R., Hancock, J., Scyphers, C., O'Gorman, T., & DiResta, R.
- Venue
- Stanford Digital Repository
- Publisher
- Stanford Digital Repository
- Published
- 2025
- OpenAlex
- W6964758640
- Cited by
- 8
- References
- 0
- Citations by year
- 2026: 3 · 2025: 2 · 2024: 3
- Type
- report
- Language
- en
- Open access
- green
- doi.org
- Locations
- Stanford Digital Repository
- Topics
- Social Sciences › Safety Research › Ethics and Social Impacts of AI
- Physical Sciences › Computer Science › Artificial Intelligence › Law, AI, and Intellectual Property
- Physical Sciences › Computer Science › Computer Vision and Pattern Recognition › Generative Adversarial Networks and Image Synthesis
- Keywords
- generative machine learning, child pornography, deepfakes, synthetic image generation, adult content, social consequences
About these checks
Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.
Statistics and arithmetic are recomputed, not judged.
Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.
What counts as a claim ›How we measure confidence ›How we check this ›
Are you an author, or do you see a problem with this page? Request a review
