Meta Adds AI to Catch Ads That Lead to Child Sexual Abuse Material

An ad does not have to display illegal material to provide a route to it. Meta is rolling out AI tools that look for seemingly harmless advertisements directing people to child sexual abuse material elsewhere online. Announced October 7, the measures also target exploitation content missed by earlier systems and people returning after account removals.

The destination becomes part of the check

The company has introduced a large language model, an AI system that processes language, to detect a tactic it calls “signposting.” These are ads suspected of steering users toward illegal content or other harmful activity outside Meta’s platforms, even when the advertisement itself appears normal.

Meta says it is assessing where an ad sends users, rather than looking only at the ad’s contents. That information can help it block destinations that violate its rules and take action against the accounts responsible. An otherwise harmless-looking ad can therefore become an enforcement target because of its suspected destination.

The company says it has recently seen bad actors use this tactic as they change their methods to evade detection. The new system is aimed at that indirect route: the advertisement can serve as a gateway to a website hosting abuse material without containing the material itself.

Finding misses and testing defenses

The ad-detection model is one component of the rollout. Other measures address material already on Meta’s services, weaknesses in its protections and the persistence of people whose accounts have been removed. Meta describes three additional enforcement efforts:

• Additional AI-driven scans seek child exploitation content that earlier systems may have missed. Meta says it will keep adding detection signals as it learns more about how these networks operate.
• A red-teaming AI agent tests the company’s own safety measures for weaknesses. Meta says this adversarial testing could reveal new methods of abuse before they become more common.
• Improved account-detection systems aim to identify people who return with new accounts after their previous accounts have been removed.

33.2 millionPieces of content acted on

Meta says it acted on 33.2 million pieces of child sexual exploitation content on Facebook and Instagram in the first half of 2026.

More than 97%Detected before user reports

The company says its systems found more than 97% of the content it acted on before users reported it.

Those figures cover Facebook and Instagram during January through June, before this October rollout. The detection percentage applies to content Meta acted on; it is not a claim that the company found that share of all exploitation content. Nor does it measure the effectiveness of the newly introduced tools.

Meta also disclosed a country-level count for India: it acted on 5.3 million pieces of child sexual exploitation content in the same six-month period. More than 98% of that content was detected before user reports, the company said.