Meta is deploying AI systems to identify advertisements that appear harmless but direct people to child sexual abuse material elsewhere online. The company calls the tactic “signposting” and says its new language-model system examines an ad’s destination, not only the content displayed on Facebook or Instagram.
That broader view can help Meta block offending websites and act against the accounts promoting them. The company is also running additional AI-driven scans to find exploitative material that earlier systems may have missed. Meta says it took action on 33.2 million pieces of child sexual exploitation content across Facebook and Instagram in the first half of 2026, with more than 97% detected before a user report.
Those figures are Meta’s own enforcement statistics and do not show how many harmful links escaped detection or how often legitimate ads were flagged. The important technical shift is from judging an isolated creative to following where it leads, which addresses an evasion method that relies on clean-looking ads as gateways. Effective deployment will require both rapid blocking and careful review because destination context can change and mistakes in this area carry serious consequences.