Student Misuse of AI-Powered “Undress” Apps

The talk provides a comprehensive and well-researched overview of the emerging issue of AI-generated CSAM, particularly in the context of student misuse. Pfefferkorn’s legal background and policy expertise lend credibility to her analysis. She effectively outlines the technical aspects of how these apps work, the legal framework, and the societal impacts, drawing on a range of sources including interviews, public records, and news reports. The argumentation is solid, with clear distinctions between different types of AI CSAM and the various stakeholders involved. However, the talk is primarily an expert opinion based on a single research paper, and some claims rely on anecdotal evidence or news reports that may not be fully verified. The speaker acknowledges limitations, such as the inability to interview school leaders, which adds to the transparency. The title accurately reflects the content, and the talk is well-structured, making it accessible to a broad audience. The Q&A session at the end provides additional insights and addresses potential concerns. Overall, the talk is a valuable contribution to the discussion on AI safety and child protection, though it would benefit from more empirical data and a broader range of perspectives.

The title accurately reflects the content, focusing on student misuse of AI-powered undress apps, which is the central theme of the talk.

8/10

The talk is based on a peer-reviewed research paper by the speaker and a colleague, involving interviews with 52 stakeholders and public records requests. The speaker is a policy fellow at Stanford HAI with legal expertise. The presentation is well-structured, cites specific laws and research, and acknowledges limitations. However, it is an expert opinion rather than a systematic review, and some claims rely on news reports and the speaker's own interpretation.

• 00:00Introduction
• 00:35Lecture
• 51:48Q&A

• Research paper on AI CSAM (by Riana Pfefferkorn and Shelby Grossman) — The talk is based on this paper, which is mentioned as published at the end of May.
• Policy brief on deepfake nudes in schools — Pfefferkorn mentions a shorter policy brief she wrote for HAI over the summer, available on the website.
• Research on Stable Diffusion 1.5 misuse — Referenced as research by a Stanford colleague (David Thiel) showing misuse of the model.
• LAION-5B dataset containing CSAM — Mentioned as a popular training dataset found to contain confirmed CSAM.
• Investigation on nudify app ecosystem — Referenced as an investigation by journalists, including a Cornell Tech professor who writes a newsletter called 'Indicator'.

• Stanford HAI policy brief on deepfake nudes — The speaker's own policy brief, which aligns with the talk's content.
• Research on Stable Diffusion 1.5 misuse — Supports the claim that open-source models can be misused for AI CSAM.