Gold Standard or Gold-Plated? Human Practices of Triple Verification in CSAM Takedown

要旨

Child sexual abuse material (CSAM) presents a critical challenge for online safety, yet the verification procedures that determine which items are classified as CSAM remain poorly understood. Triple verification (requiring three reviewers to agree) is promoted as a safeguard, but little is known about how it is implemented, how it is perceived by experts, and how voting conditions affect reliability. We address this gap through a mixed-methods study. We interviewed 14 experts from seven organizations (e.g., law enforcement, hotlines, etc.) to map current verification practices, then ran an inter-reliability experiment with Dutch National Police experts who reviewed 2,031 images and videos under different voting conditions (blind vs. non-blind, varied order). Finally, we held a focus group to explore the reasons behind disagreements. We find that practices vary widely, perceptions of triple verification reflect both safeguards and burdens, and expert agreement depends on voting conditions and content type.

受賞
Honorable Mention
著者
Melissa Rottier
Delft University of Technology, Delft, Netherlands
Michel van Eeten
Delft University of Technology, Delft, Netherlands
Savvas Zannettou
Delft University of Technology, Delft, Netherlands

会議: CHI 2026

ACM CHI Conference on Human Factors in Computing Systems

セッション: Community Governance and Moderation

P1 - Room 114
7 件の発表
2026-04-17 18:00:00
2026-04-17 19:30:00