Polls | Gaza | Fact Checker | Support Us
Sign In | Join Free |
UK FACT CHECK POLITICS
UK FACT CHECK POLITICS
 Voting intention Labour Lab 26% Reform Ref 21% Con Con 20% Green Grn 14% Lib Dem LD 10% Poll tracker

Menu

Topics
About
Get Involved
Follow Us

Account

AI-Generated Child Sexual Abuse Material: Insights from Educators, Platforms, Law Enforcement, Legislators, and Victims

purl.stanford.edu 05 September 2026 at 05:09 View original article →

The original link is the material being checked. It may contain false or misleading claims — we link for transparency, not endorsement.

92/100
Verdict High Trust Highly Credible — Primary Institutional Research Record (Verified Abstract/Metadata)

The submitted text is not journalism or commentary but a verbatim reproduction of the abstract and Stanford Digital Repository metadata for the report 'AI-Generated Child Sexual Abuse Material: Insights from Educators, Platforms, Law Enforcement, Legislators, and Victims' (Grossman, Pfefferkorn & Liu, 2025), DOI 10.25740/mn692xc5736.

✓ 9 checks out ✕ 0 disputed ? 5 unverified
☰ Full analysis The complete summary ⌄

The submitted text is not journalism or commentary but a verbatim reproduction of the abstract and Stanford Digital Repository metadata for the report 'AI-Generated Child Sexual Abuse Material: Insights from Educators, Platforms, Law Enforcement, Legislators, and Victims' (Grossman, Pfefferkorn & Liu, 2025), DOI 10.25740/mn692xc5736. Every substantive descriptive and methodological claim in the excerpt was matched word-for-word against the primary record at https://purl.stanford.edu/mn692xc5736, the full PDF at https://stacks.stanford.edu/file/mn692xc5736/AI-CSAM-paper-2025-05-29.pdf, and the publisher's announcement at https://cyber.fsi.stanford.edu/news/ai-csam-report. Methodology figures (52 interviews, documents from four public school districts, coded state legislation) are confirmed by three independent Stanford-hosted pages, with the HAI policy brief adding that interviews ran from mid-2024 to early 2025. The four headline findings — unclear school prevalence and inadequate school prevention, platforms not systematically flagging AI-generated status in NCMEC CyberTipline reports, frontline staff perceiving low platform prevalence, and legal risk impeding CSAM red teaming — are faithfully reproduced; the red-teaming finding is independently echoed in Tech Policy Press (26 September 2025) quoting co-author Pfefferkorn. Two caveats reduce the score below the mid-90s. First, the harms framing in the opening paragraph rests partly on a contested causal proposition (that viewing AI CSAM normalises abuse and raises contact-offending risk); the report itself hedges this appropriately as what 'many experts warn', and bodies such as the IWF assert it, but robust causal evidence remains limited, so it is adjudicated as a correctly attributed but not independently established claim. Second, the excerpt is a snapshot of a 2024–early-2025 evidence base: the 'prevalence remains low' perception is time-bound and sits uneasily against later NCMEC data reporting more than 1.5 million CyberTipline reports with a generative-AI nexus in 2025. Minor metadata artefacts (truncated DOI and Location fields rendered as '[', an author-order discrepancy between the metadata block and the preferred citation, and an unconfirmed 'Version 1, Jul 7, 2025' stamp) are formatting or ingestion issues rather than accuracy failures, and the omission of the Safe Online funding acknowledgement — which appears on the publisher's page — is a transparency gap in the excerpt as supplied.

What checked out (9)
  • VERIFIED — Bibliographic identity: the report 'AI-Generated Child Sexual Abuse Material: Insights from Educators, Platforms, Law Enforcement, Legislators, and Victims' by Grossman, S., Pfefferkorn, R. and Liu, S. (2025) exists and is deposited in the Stanford Digital Repository under DOI 10.25740/mn692xc5736 (primary: https://purl.stanford.edu/mn692xc5736).
  • VERIFIED — Abstract text is verbatim: the harms paragraph and findings paragraph in the submission match the repository abstract and the full PDF character-for-character (https://purl.stanford.edu/mn692xc5736/version/1; https://stacks.stanford.edu/file/mn692xc5736/AI-CSAM-paper-2025-05-29.pdf).
  • VERIFIED — Methodology: '52 people' interviewed, 'documents from four public school districts' analysed, and state legislation coded, confirmed on three independent Stanford-hosted pages (https://cyber.fsi.stanford.edu/news/ai-csam-report; https://purl.stanford.edu/mn692xc5736; https://hai.stanford.edu/assets/files/hai-policy-brief-addressing-ai-csam.pdf).
  • VERIFIED — Fieldwork window: the Stanford HAI policy brief (July 2025) states the 52 interviews were conducted between mid-2024 and early 2025, corroborating the study period implied by the excerpt (https://hai.stanford.edu/assets/files/hai-policy-brief-addressing-ai-csam.pdf).
  • VERIFIED — Schools finding: prevalence of student-on-student nudify app use is unclear, schools are generally not addressing nudify risks with students, and some schools mishandled incidents; reproduced accurately and repeated in the HAI brief's key takeaways (https://cyber.fsi.stanford.edu/news/ai-csam-report; https://hai.stanford.edu/policy/addressing-ai-generated-child-sexual-abuse-material-opportunities-for-educational-policy).
  • VERIFIED — CyberTipline finding: mainstream platforms report discovered CSAM without systematically discerning or conveying AI-generated status, shifting identification to NCMEC and law enforcement (https://cyber.fsi.stanford.edu/news/ai-csam-report; https://purl.stanford.edu/mn692xc5736/version/1).
  • VERIFIED — Red-teaming finding: legal risk hinders CSAM red teaming at mainstream AI model-building companies; independently corroborated by Tech Policy Press (26 September 2025), which quotes co-author Pfefferkorn on companies facing legal exposure when testing models, and notes state-level adversarial-testing carve-outs do not shield against federal law (https://www.techpolicy.press/how-congress-could-stifle-the-onslaught-of-aigenerated-child-sexual-abuse-material/).
  • VERIFIED — Licence: the deposit is released under a Creative Commons Attribution 4.0 International licence (CC BY), as stated on the repository record (https://purl.stanford.edu/mn692xc5736/version/1).
  • VERIFIED — Publication dating: the 28/29 May 2025 dates are consistent with the repository record (dated 29 May 2025) and the PDF filename and stamp 'AI-CSAM-paper-2025-05-29' (28 May 2025) (https://stacks.stanford.edu/file/mn692xc5736/AI-CSAM-paper-2025-05-29.pdf).
? Unverified claims 5 claims ⌄
  • UNVERIFIED — 'Research team head: Hancock, Jeffrey'. Jeffrey Hancock is confirmed as a Stanford Cyber Policy Center co-director and Social Media Lab principal (https://cyber.fsi.stanford.edu/people/jeff-hancock; https://sml.stanford.edu/people/jeff-hancock), but no retrieved source explicitly confirms his designation as research team head for this specific deposit. Treated as plausible repository metadata, not independently confirmed; not marked False.
  • UNVERIFIED — 'Version 1 | Jul 7, 2025'. Retrieved snapshots of the repository record confirm a versioned URL structure and Version 1 citation, but the 7 July 2025 version timestamp was not visible in retrieved content. Likely a routine deposit/version-update date distinct from the May 2025 publication date; insufficient evidence to confirm or refute.
  • UNVERIFIED (CONTESTED UNDERLYING SCIENCE) — The proposition that viewing AI CSAM normalises child abuse and increases contact-abuse risk. The excerpt correctly attributes this as an expert warning rather than asserting it as settled fact, and it is asserted by the Internet Watch Foundation (https://www.iwf.org.uk/about-us/why-we-exist/our-research/how-ai-is-being-abused-to-create-child-sexual-abuse-imagery/) and argued in peer-reviewed/preprint literature (https://arxiv.org/abs/2510.02978, which describes AI CSAM as a possible pathway into offending through desensitisation and lowered barriers). However, the causal claim rests on theory and practitioner observation rather than established causal evidence, and the same literature acknowledges an active dispute about AI CSAM's harm profile. The report's hedged phrasing is accurate; the underlying causal link is not verified.
  • UNVERIFIED — Law-enforcement resource-diversion claim (officers mistaking AI CSAM for a real unidentified victim and wasting victim-identification effort). Widely reported as a practitioner concern and consistent with the report's interview basis, but no retrieved primary dataset quantifies the diversion. Directionally supported, not quantitatively established.
  • UNVERIFIED — Truncated metadata fields 'DOI | [' and 'Location | [' and the empty preferred-citation URL. These are ingestion/rendering artefacts in the submitted text; the underlying DOI resolves correctly (10.25740/mn692xc5736), so the artefacts are formatting defects rather than factual errors.
⌕ Sources & how we checked Search journal, source grades, confidence ⌄
Confidence

High (0.88) — Confidence is high because the submission is a verbatim reproduction of a primary, DOI-registered institutional record that was retrieved directly and matched line-by-line, with methodology figures triangulated across three independent Stanford-hosted pages and at least one substantive finding corroborated by a non-Stanford outlet quoting a co-author. Confidence is held below 0.95 for four reasons: the underlying interview data are non-public and therefore unauditable, so the study's empirical findings are accepted on the authors' representation rather than replication; most confirmatory sources are institutionally affiliated with the depositor; three metadata and attribution details (Hancock's project role, the Version 1 timestamp, the truncated fields) could not be resolved from retrieved content; and the excerpt's prevalence-related statements are a time-bound snapshot whose present-day currency is materially affected by later NCMEC and IWF reporting. No evidence was found that would justify a False adjudication on any claim, and all residual uncertainty has been recorded as Unverified with stated limits rather than resolved by assumption.

Search journal

Stanford AI-Generated Child Sexual Abuse Material Insights from Educators Platforms Law Enforcement Legislators Victims

Grossman Pfefferkorn Liu 2025 AI CSAM report Stanford Internet Observatory 52 interviews

Stanford Digital Repository AI-generated CSAM report May 2025 nudify apps schools

"52" interviews AI CSAM Stanford report four school districts state legislation coded

Stanford HAI policy brief addressing AI-generated CSAM schools nudify key takeaways

platforms NCMEC CyberTipline reports do not indicate AI-generated Stanford 2025 red teaming legal risk

Jeff Hancock Shelby Grossman Riana Pfefferkorn Sunny Liu AI CSAM Stanford Cyber Policy Center

doi 10.25740/mn692xc5736 AI-Generated Child Sexual Abuse Material

Sunny Liu Jeffrey Hancock Stanford Social Media Lab researcher

IWF NCMEC AI-generated CSAM reports diverting law enforcement resources real victim identification 2025

Safe Online funding Stanford AI CSAM research grant 2025

news coverage Stanford study schools nudify apps deepfake nudes students response missteps 2025

evidence debate whether viewing AI-generated CSAM increases contact offending contested researchers

legal risk red teaming CSAM AI models companies cannot test child safety law barrier 2025

Article metrics

Emotion 13% · Reading grade 13.3 · 447 words

≡ The article we checked Full text as retrieved ⌄
Abstract AI-generated child sexual abuse material (CSAM) carries unique harms. When generated from a photo of a clothed person, it can damage that person’s reputation and cause serious distress. When based on existing CSAM, it risks re-traumatizing victims. Even AI CSAM that seems purely synthetic may come from a model that was trained on real abusive material. Many experts also warn that viewing AI CSAM can normalize child abuse and increase the risk of contact abuse. There is the added risk that law enforcement may mistake AI CSAM for content involving a real, unidentified victim, leading to wasted time and resources spent trying to locate a child who does not exist.In this report we aim to understand how educators, platform staff, law enforcement officers, U.S. legislators, and victims are thinking about and responding to AI CSAM. We interviewed 52 people, analyzed documents from four public school districts, and coded state legislation.Our main findings are that while the prevalence of student-on-student nudify app use in schools is unclear, schools are generally not addressing the risks of nudify apps with students, and some schools that have had a nudify incident have made missteps in their response. We additionally find that mainstream platforms report the CSAM they discover, but, for various reasons, without systematically trying to discern and convey whether it is AI-generated in their reports to the National Center for Missing and Exploited Children’s (NCMEC) CyberTipline. This means the task of identifying AI-generated material falls to NCMEC and law enforcement. However, frontline platform staff believe the prevalence of AI CSAM on their platforms remains low. Finally, we find that legal risk is hindering CSAM red teaming efforts for mainstream AI model-building companies. | Resource Type | text | | | | Publication date | May 29, 2025; May 28, 2025 | | Author | Pfefferkorn, Riana | | | | | Author | Grossman, Shelby | | | Author | Liu, Sunny | | | Research team head | Hancock, Jeffrey | | | Subject | Artificial intelligence | | | | Subject | AI (Artificial intelligence) | | Subject | Intelligence, Artificial | | Subject | Child sexual abuse | | Genre | Text (Report) | | DOI | [ | | | | Location | [ | Use and reproduction User agrees that, where applicable, content will not be used to identify or to otherwise infringe the privacy or confidentiality rights of individuals. Content distributed via the Stanford Digital Repository may be subject to additional license and use restrictions applied by the depositor. License This work is licensed under a Creative Commons Attribution 4.0 International license (CC BY). Preferred citation Grossman, S., Pfefferkorn, R., & Liu, S. (2025). AI-Generated Child Sexual Abuse Material: Insights from Educators, Platforms, Law Enforcement, Legislators, and Victims. Version 1. Stanford Digital Repository. Available at | Version 1 | Jul 7, 2025 | You are viewing this version | [Copy URL]( | | | Each version has a distinct URL, but you can use this PURL to access the latest version. [ Loading usage metrics...

Share this fact check

← Check another article or image