Meta has deployed new AI tools to detect ads that look harmless but covertly direct people to child sexual abuse material elsewhere online. In its October 7 announcement, the company says it widened an investigation into the scheme, disabled violating accounts and blocked the external links involved. TechCrunch reports that the tools target both ads and accounts used to direct people to illegal material.
Ad review follows the destination
The new large language model system looks for what Meta calls “signposting.” These are seemingly benign ads that the company strongly suspects of directing people to illegal content or harmful activity outside its apps. Meta says it has improved its understanding of where an ad leads so it can block violating destinations and act against the accounts responsible.
Meta is also running additional AI scans for child sexual exploitation material that earlier systems missed. It says it will expand the signals those scans use as its investigation progresses. A red-teaming AI agent probes the company’s defenses for weaknesses, while stronger account detection aims to catch offenders who return after removal.
Destination checks already have an enforcement mechanism. Meta says that once it blocks a link, it searches for and deletes ads, posts and comments containing it. Ads containing blocked links face rejection at upload. The company also uses PhotoDNA and other matching tools to detect known abuse images and videos, including in external link previews.
What Meta’s figures measure
Meta says it acted on 33.2 million pieces of child sexual exploitation content on Facebook and Instagram between January and June 2026. Its systems found and addressed more than 97% before anyone reported them. The announcement gives no separate accuracy results for the new signposting system.
