Meta rolls out new AI tools to detect ads that secretly lead to child sexual abuse material

Meta has introduced new AI tools to identify advertisements that covertly direct users to child sexual abuse material on its platforms. The company disclosed that it removed 33.2 million pieces of such content from Facebook and Instagram during the first half of 2026. Over 97% of the deleted content was automatically detected by Meta's systems prior to user reports.

In India, Meta removed 5.3 million pieces of child sexual exploitation content during the same period, with more than 98% of it identified before anyone reported it. These new AI tools focus on detecting "signposting" in ads, which refers to ads that appear harmless but are suspected of leading users to illegal or harmful content elsewhere online.

Meta has developed a new large language model (LLM) system to identify this type of signposting, as bad actors have recently evolved their methods to avoid detection. The company will now examine where ads direct users, not just the content of the ads themselves, to block websites or accounts that violate its rules. Additionally, Meta is using advanced AI scans to find hidden child exploitation content that previous systems might have missed.

A "red-teaming AI agent" is being used to test Meta's safety measures and uncover any weaknesses that could be exploited by bad actors. The company is also enhancing its ability to detect users who create new accounts after their previous accounts have been removed. These measures come as Meta faces increasing scrutiny over child safety on its platforms, following lawsuits and criticism from lawmakers regarding the potential risks its services pose to young users.