In India, Meta said it acted on 5.3 million pieces of child sexual exploitation content between January and June 2026, and that more than 98% was detected before users reported it. The company also said it acted on 33.2 million pieces of that content on Facebook and Instagram in the first half of 2026, with more than 97% found by its systems before user reports.

Meta says the large language model system targets signposting, where an ad may appear harmless but point users toward illegal material elsewhere. Destination signals support blocking websites or other destinations that break its rules, as well as action against the accounts behind them.

The announcement does not name the model, give a parameter count, benchmark, latency, or measure how many ads, accounts, or destinations were removed. It also does not state the LLM's accuracy, cost, or share of the 5.3 million actions.

Meta also described additional AI scans to catch child exploitation content earlier systems may have missed, a red-teaming AI agent that tests its own safety measures, and improved systems for finding people who return with new accounts after prior removals.

The measures arrive while Meta faces pressure over child safety, including lawsuits and criticism from lawmakers. In August, it reached an agreement to pay up to $18 billion to settle a child safety lawsuit involving 29 U.S. states.

The stronger lead is destination review, not the LLM itself, because the source ties the new tool to signposting and enforcement scope rather than to a named model or measured removals.