Meta has introduced additional artificial intelligence tools to strengthen its advertising review systems and identify advertisements that may direct users to child sexual exploitation material hosted outside its platforms.
The update follows the company's identification of tactics in which seemingly harmless advertisements contain indirect references or signals leading users to prohibited content on external websites.
To address this, Meta has incorporated large language models (LLMs) into its detection systems to identify what it calls "signposting". The technique involves using apparently ordinary advertising content to guide users towards harmful or illegal material elsewhere online.
The company has also expanded its ad review process to examine the destinations linked to advertisements, rather than evaluating only the content displayed within them. This allows its systems to identify potentially violating websites and investigate accounts associated with such activity.
Meta said it has introduced additional AI-powered reviews of advertising content to identify material that earlier detection systems may have missed.
A new AI red-teaming agent has also been deployed to test the effectiveness of its detection mechanisms. The system is designed to identify weaknesses and emerging methods used to evade advertising safeguards.
Alongside these measures, Meta has strengthened its ability to detect previously removed users attempting to create new accounts.
The company said its broader detection infrastructure combines behavioural analysis, AI models and image-matching technologies, including PhotoDNA, to identify suspected child sexual abuse material (CSAM).
Meta also participates in the Tech Coalition's Lantern programme, through which companies share information about potentially harmful activity, including digital fingerprints of identified images and videos.
When Meta blocks links associated with violating content, it searches for the same links across advertisements, posts and comments. The company also takes measures to prevent blocked links from being shared on Facebook, Instagram and Threads.
Between January and June 2026, Meta took action against 33.2 million pieces of child sexual exploitation content across Facebook and Instagram globally. According to the company, more than 97% of the content was detected and addressed before users reported it.
In India, Meta reported taking action against 5.3 million pieces of such content during the same period, with more than 98% identified before receiving user reports.
The additional safeguards follow scrutiny of Meta's advertising systems. An investigation by the Tech Transparency Project identified 332 paid advertisements containing child sexual abuse material on Facebook and Instagram in 2026, including allegedly AI-manipulated images.
Responding to the findings, Meta said it does not tolerate child exploitation involving real or AI-generated content. The company stated that many flagged advertisements had already been removed, most received fewer than 200 impressions, and their combined advertising expenditure was below $5,000.
The latest measures expand Meta's use of AI in advertising moderation, with greater attention to external destinations, account behaviour and attempts to circumvent existing safety controls.
Disclaimer: This article may include information derived from interviews, press releases, public statements, research, company communications and other publicly available or third-party sources. Such material may be summarised, paraphrased or contextualised for journalistic and editorial purposes. All rights in third-party content remain with their respective owners.
Related Articles
