AI image content moderation guardrails silently fail when platforms update content policies, new research from Fudan ...