Meta said on October 7, 2026, that it had deployed additional AI measures to detect seemingly benign ads suspected of directing people from Facebook or Instagram to illegal content or other harmful activity elsewhere. The measures include a large language model (LLM) system focused on what Meta calls “signposting.”
Meta’s system looks beyond an ad’s content
Meta describes signposting as suspected covert redirection: an ad may look ordinary while leading people to harmful or illegal material off-platform. The new LLM system targets that behavior, while Meta’s ad review considers where an ad leads as well as what it shows. Meta says it can block violating destinations and take action against the accounts responsible.
The suspected redirection is the focus; Meta does not say that every ad flagged this way contains illegal material.
Other announced measures target missed content and repeat accounts
Meta also described additional AI-driven scans intended to find child sexual exploitation material in ads that earlier systems missed. A separate AI red-teaming agent probes Meta’s defenses for weaknesses, with the aim of surfacing new tactics before they spread. The company said it strengthened detection of people who try to create new accounts after an earlier account is removed.
What Meta reported for January–June 2026
For January–June 2026, Meta said it took action on 33.2 million pieces of child sexual exploitation content across Facebook and Instagram worldwide. More than 97% were found and addressed before a user report. In India, Meta reported 5.3 million pieces of such content actioned during the same period, with more than 98% found and addressed before a report.
Those figures count content Meta acted on across the platforms; the new signposting system targets suspected ad redirection.
