Meta announced a set of new artificial intelligence tools targeting tactics in which ads do not clearly appear to violate the rules but are used to direct users to external websites hosting child sexual abuse material. In parallel, the company said it took action against 33.2 million items related to child sexual exploitation on Facebook and Instagram during the first half of 2026, after its systems detected more than 97% of them before users reported them.
In India, Meta handled 5.3 million items during the same period and said that more than 98% of them were detected proactively. The figures in the source do not specify the number of accounts or ads associated with this material, but they reflect the company’s growing reliance on automated detection before human reports are submitted.
Detecting the Destination, Not Just the Ad
The most prominent new tool uses a system based on a large language model to detect what Meta calls “signposting.” The term refers to ads that may appear ordinary but are suspected of being used as gateways to illegal content or harmful activity outside the company’s platforms.
Under the new approach, Meta does not limit itself to analyzing the ad’s content; it also examines the destination to which the user is sent. If the site or destination is found to violate the company’s rules, Meta can block it and take action against the accounts behind it. Meta says the change followed observations that abusive actors were developing tactics to evade traditional detection tools.
Red-Teaming and Tracking Returning Accounts
The company is also using additional AI-powered scanning operations to find material that its previous systems may not have detected, adding new signals as it gains a better understanding of how these networks operate.
Meta has also developed a “red-teaming AI agent” tasked with searching for weaknesses in internal protection systems and simulating the methods abusers might use to bypass them. The company also announced improvements to its ability to detect people who return to the platforms with new accounts after their previous accounts have been removed.
Why Does This Matter?
The practical change lies in shifting part of safety monitoring from analyzing a post or ad to analyzing the complete path that may begin with an ad that appears lawful and end at a harmful external website. This broadens the scope of preventive intervention, but it makes the system’s effectiveness dependent on its ability to assess external destinations and relationships between accounts—areas that may require continuous updating as abusive networks change their tactics.
The measures come amid ongoing legal and political pressure on Meta over child protection. In August, the company reached an agreement to pay up to 18 billion dollars to settle a child safety lawsuit filed by 29 U.S. states. During the year, it also launched features including parental control tools for Meta AI, accounts designed for preteens on WhatsApp, and alerts for parents when children search Instagram for content related to self-harm.
certi.news analysis: The announcement represents a clear expansion of AI’s role, from detecting violating content to analyzing the structure that leads to it. However, the source does not provide independent measurements of the accuracy of the “signposting” tools, nor does it specify the number of sites or accounts blocked as a result of the new tools. The quantitative results announced therefore remain tied to Meta’s own reports.