Meta rolls out new AI tools to detect ads that secretly lead to child sexual abuse material
techcrunch ·
Meta announced that it removed 33.2 million pieces of child sexual exploitation content from Facebook and Instagram during the first half of 2026, with over 97% detected by its systems before user reports. To combat evolving threats, the company is deploying new AI tools designed to identify ads that act as gateways to illegal child abuse material hosted outside its platforms. A key innovation is a large language model system that detects "signposting," where harmless-looking ads secretly direct users to harmful websites. Meta is now analyzing ad destinations rather than just ad content to block rule-breaking sites and take action against associated accounts. Additional measures include red-teaming AI agents that test safety defenses for weaknesses and improved systems for tracking users who create new accounts after previous bans. These efforts come amid significant legal pressure, including an $18 billion settlement agreement with 29 U.S. states regarding child safety concerns. The company has also introduced various parental control features across its apps, such as alerts for self-harm searches on Instagram and enhanced privacy settings on WhatsApp.
techcrunch