Meta Adds AI Systems to Detect Ads Pointing to Child Abuse Material
Meta expands child-safety enforcement with AI
Meta says it acted on 33.2 million pieces of child sexual exploitation content across Facebook and Instagram during the first half of 2026. More than 97% of the material was identified by its systems before any user report, the company said. In India, Meta acted on 5.3 million pieces, with proactive detection exceeding 98%.
Alongside those figures, Meta is rolling out AI tools intended to identify ads and accounts that may be used to steer people towards illegal child sexual abuse material outside its platforms. The measures address tactics in which an advertisement appears innocuous on its face but functions as a route to prohibited material hosted elsewhere online.
Looking beyond the advertisement itself
The company has introduced a large language model system to detect what it calls “signposting.” Meta uses the term for ads suspected of directing users to illegal content or other harmful activity beyond its services. The ads may not themselves contain illegal material, but they can act as gateways to sites that do.
Meta said it has recently observed bad actors adapting their methods to avoid detection. Its new approach therefore considers the destination to which an advertisement sends a user, rather than assessing only the advertisement’s visible content. That information can be used to block websites and other destinations that violate Meta’s rules, and to take action against the accounts associated with them.
The company is also deploying additional AI-driven scans to identify child exploitation content that earlier systems may have missed. Meta said it plans to add signals as it develops a better understanding of how such networks operate. This work complements measures such as Meta AI-chat parental safety alerts, which focus on parental alerts and escalations when AI-chat interactions indicate self-harm risk.
Testing defences and addressing repeat accounts
Another tool is a “red-teaming AI agent” designed to test Meta’s own safety protections. It seeks weaknesses that could allow bad actors to circumvent those defences, with the aim of identifying emerging methods of abuse before they become more widespread.
Meta is also improving its ability to find people who return to its platforms with new accounts after earlier accounts have been removed. The announcement arrives amid continuing scrutiny of child safety on Meta’s services. In August, Meta agreed to pay up to $18 billion to settle a child-safety lawsuit involving 29 US states.
For businesses operating digital services, Meta’s approach underlines a practical requirement: trust-and-safety controls need to examine destinations, linked accounts and attempts to evade enforcement, not solely the content that is immediately visible to users.