They found a way to spot AI models built to create child abuse imagery—without ever making a single picture. The accuracy is staggering, and it’s already changing how platforms fight back. But the real shock is how fast this threat has grown.