To address that behavior, Meta has deployed a new large language model for detecting what it calls “signposting.” The system is intended to identify seemingly benign advertising that may covertly guide users toward child sexual exploitation material or other prohibited activity outside Meta’s services.
The company is also analyzing an ad’s destination rather than evaluating only its visible content. When Meta identifies an external site that violates its rules, it can block the destination and take enforcement action against associated accounts.
Meta said it is conducting additional AI-based scans of advertising to uncover material missed by earlier detection systems. The company plans to continue incorporating new signals as its investigations reveal changes in how offenders operate.
Another addition is an AI red-teaming agent designed to test Meta’s own defenses. The system probes for vulnerabilities that could be exploited to evade detection, with the goal of identifying emerging methods before they become more widespread.
Meta has also strengthened systems intended to identify repeat offenders who create new accounts after earlier ones have been removed.
The new tools add to the company’s existing automated systems. Meta uses behavioral signals to identify suspicious account activity and combines classifiers, AI models and other indicators to detect emerging patterns of abuse.
Its moderation infrastructure also includes PhotoDNA and other image- and video-matching technology. Meta has used those systems across its apps since 2011 to find identical or similar copies of known child sexual abuse material.
When Meta identifies such material, the company says it removes the content and may disable accounts when evidence indicates malicious distribution. It also reports qualifying material through the National Center for Missing & Exploited Children in accordance with applicable law.
Meta maintains a database of image and video hashes associated with known material and contributes new hashes for sharing with other companies through the Tech Coalition’s Lantern program. The same matching technology is used to examine previews of external links.
The company also blocks links pointing to websites that host or generate prohibited material. After blocking a destination, Meta says it searches for other ads, posts or comments containing that link and removes them. Its systems are also designed to prevent blocked links from being posted across Facebook, Instagram and Threads.
The scale of Meta’s enforcement remains significant. From January through June, the company said it acted on 33.2 million pieces of child sexual exploitation content worldwide on Facebook and Instagram. More than 97% was found proactively.
In India, Meta acted on 5.3 million pieces during the same period, with more than 98% identified before a user report.
Meta said its approach also includes manual investigations by specialists, including former FBI investigators, who examine networks of accounts rather than limiting enforcement to individual profiles.
“Sophisticated bad actors who try to exploit children online don’t stay in one place,” Meta said. “These determined predators move across different platforms, and change their tactics to evade detection.”
The company is also expanding cooperation with authorities. Meta said in September that it would report child safety cases directly to India’s National Cyber Crime Reporting Portal, building on its existing reporting process involving the National Center for Missing & Exploited Children.
The safety push follows other child-focused changes Meta has introduced this year across its services. Those include parental controls for Meta AI, parent-linked WhatsApp accounts for younger users, Instagram alerts related to searches for self-harm content and additional controls for teen accounts on WhatsApp.
Meta continues to face scrutiny over child safety across its platforms. In August, the company agreed to pay up to $18 billion to settle a lawsuit involving 29 U.S. states over alleged harms to children.
The latest AI systems are aimed at a narrower technical problem: detecting abuse even when prohibited material itself never appears directly inside an ad. By examining destinations, account behavior and emerging evasion techniques together, Meta is trying to identify networks that use its platforms as an entry point to illegal activity elsewhere online.
This analysis is based on reporting from Meta & TechCrunch.
Image courtesy of Meta.
This article was generated with AI assistance and reviewed for accuracy and quality.
