Meta is expanding its use of artificial intelligence to identify accounts and advertisements that may be used to secretly direct people toward illegal child sexual exploitation material outside its platforms.
The company said it acted against 33.2 million pieces of child sexual exploitation content on Facebook and Instagram during the first half of 2026. More than 97% of the content was detected by Meta’s systems before it was reported by users.
In India, Meta said it took action against 5.3 million pieces of such content, with more than 98% detected before user reports.
The company says its latest measures are designed to address a changing tactic in which seemingly harmless advertisements can be used as a gateway to illegal material elsewhere online.
What Is Meta Changing?
Meta has introduced a new large language model-based system designed to identify what it calls “signposting.”
In this context, signposting refers to advertisements that may appear normal on the surface but could be directing users toward illegal or harmful destinations outside Meta’s services.
The company says it has recently seen bad actors change their methods to avoid traditional detection systems. As a result, Meta is looking beyond the content of an advertisement itself.
The company can now analyze where an advertisement sends users and use that information to identify potentially harmful destinations.
If a destination violates Meta’s policies, the company can block it and take action against the accounts associated with it.
AI Systems Will Look for Content Earlier Systems Miss
Meta is also deploying additional AI-based scanning systems to identify child exploitation content that may have escaped its earlier detection technologies.
The company says it will continue adding new signals to its systems as it learns more about how networks involved in child exploitation operate.
The goal is to make it harder for people involved in these activities to adapt their behavior around existing automated safeguards.
Meta Uses an AI Agent to Test Its Own Safety Systems
Another part of the new approach involves a red-teaming AI agent.
The system is designed to test Meta’s own safety protections by looking for weaknesses that could potentially be exploited to bypass them.
This type of testing allows Meta to simulate potential attempts to get around its safeguards and identify gaps before they become more widely used.
The company says the approach could help it discover new forms of abuse and strengthen its defenses more quickly.
Tracking Repeat Offenders
Meta is also improving its ability to identify people who return to its platforms after their previous accounts have been removed for violating its rules.
The company says these systems are intended to help prevent repeat offenders from simply creating new accounts and continuing prohibited activity.
Meta Faces Continued Pressure Over Child Safety
The new AI measures come as Meta continues to face scrutiny over how it protects children across its social media platforms.
The company has faced lawsuits, criticism from lawmakers and broader concerns about the potential risks social media services can pose to young users.
In August, Meta agreed to pay up to $18 billion to settle a child-safety lawsuit involving 29 U.S. states. The case focused on allegations and concerns surrounding the impact of social media platforms on children.
Meta has also introduced several child-safety measures during 2026.
These include parental controls for Meta AI, parent-linked accounts for preteens on WhatsApp and alerts for parents when teenagers search Instagram for certain self-harm-related content.
WhatsApp also added additional parental controls in September, giving parents more options for managing how teenagers use Channels, who can view their status updates and who can add them to groups.
Why It Matters
The latest changes highlight how online safety companies are increasingly using AI to detect not only prohibited content but also the behavior and networks surrounding it.
For Meta, simply identifying illegal content inside its platforms is no longer the only challenge. The company also needs to identify attempts to use its services as a pathway to harmful activity elsewhere on the internet.
Using AI to analyze advertisements, destinations, account behavior and potential weaknesses in its own safety systems could give Meta more ways to identify these activities earlier.
However, the effectiveness of these systems will depend on how accurately they can detect evolving abuse patterns while avoiding unnecessary action against legitimate users and advertisers.
Conclusion
Meta says it is expanding its AI-based safety systems to stay ahead of people who attempt to misuse Facebook, Instagram and other services to facilitate child exploitation.
The new tools focus on several areas, including suspicious advertising patterns, external destinations, missed content, repeat offenders and weaknesses in Meta’s own safety systems.
As online threats continue to change, Meta’s latest approach shows the growing role AI is playing in automated platform safety and the company’s ongoing effort to strengthen protections for children.