Position: Preventing AI-Generated CSAM Necessitates New Approaches to AI Safety

Existing AI safety methods are inadequate for child protection: Standard safety techniques like dataset auditing, red teaming, and fine-tuning prevention cannot be applied to AI-generated child sexual abuse material (CSAM) because ethical and legal constraints prevent researchers from accessing, examining, or testing with such material—creating a critical gap in how we can evaluate and improve AI safety in this domain.

AI is actively being misused to harm children: The paper documents that modern AI systems are already being exploited to create AI-generated CSAM, facilitate child sexual exploitation, and lower barriers to abuse, indicating this is not a theoretical future risk but a present problem requiring urgent attention.

Child protection requires rethinking AI development practices: The paper identifies 15 open problems spanning the entire AI lifecycle—from how training data is collected and models are designed to how systems are deployed and maintained—suggesting that protecting children demands fundamental changes to how AI systems are built, not just add-on safety features.

Policymakers and developers need tailored guidance: The authors call for targeted recommendations specific to child safety rather than generic responsible AI principles, recognizing that preventing AI-facilitated child abuse requires specialized approaches that differ from other AI safety challenges.

This should be treated as a core safety priority: The paper argues that preventing AI-facilitated child sexual abuse must become a "central, safety-critical dimension" of AI research itself, not a peripheral concern, requiring dedicated resources and expertise comparable to other major AI safety challenges.