On a recent Lawfare podcast, Kevin Frazier spoke with Dave Willner and Vaishnavi J. about how AI platforms handle child safety. They argue that common responses—age gates, parental consent, and bans—are blunt tools that don't scale well.
The guests have released open-source policies for teen self-harm and suicidal ideation content. The policies use age-banded rules and a four-zone framework: remove content, suppress it with a warning, suppress it from recommendations, or allow it. This is meant to give platforms more granular options than a simple allow/block binary.
The policies were built using more than 1,200 hand-labeled examples. Willner and Vaishnavi also describe how Zentropi's CoPE classifier turns written policy into enforcement, and they discuss how platforms should weigh false positives, false negatives, over-refusal, under-refusal, human review, and appeals.