Meta is expanding AI-powered detection beyond individual posts to identify covert adverts, off-platform links, repeat offenders and account networks linked to child sexual exploitation.
Meta is deepening its use of artificial intelligence to counter child sexual exploitation across Facebook, Instagram and Threads, as it moves towards a more aggressive, network-based enforcement model. According to the company, the latest systems are intended to detect increasingly elaborate attempts to evade moderation, including activity routed through external websites and advertising that appears benign at first glance. Meta is focusing in particular on malicious ads that can function as covert signposts to illegal material hosted away from its own platforms.
A central part of the update is an LLM-based detector designed to identify those hidden signposts. Meta says the tool is built to recognise advertising that may not contain overtly abusive content, but that nonetheless appears to guide users towards harmful destinations. The company has also improved systems that examine where an advert leads, not merely what it shows, so that links connected to violating material can be blocked and the accounts distributing them can face enforcement action. In parallel, Meta says it is widening automated sweeps of ad content to uncover material that previous systems may have missed.
The company is also leaning more heavily on adversarial testing. Meta says it has deployed a red-teaming AI agent that simulates attacker behaviour in order to probe weaknesses in its own defences. That matters because child exploitation networks often adapt quickly, shifting language, account structures and distribution methods to avoid detection. Meta says it is pairing that approach with stronger recidivism detection, so that people who return after account removal can be identified more effectively. The company is also increasingly investigating clusters of accounts rather than treating each violation as an isolated case, a strategy that reflects the way coordinated abuse often operates in practice.
Meta says the new systems sit alongside older tools that remain central to its enforcement stack. The company has used PhotoDNA and other image- and video-matching technologies since 2011, and says it maintains one of the industry’s largest collections of image and video hashes for this purpose. It also scans off-platform link previews, blocks links known to lead to offending websites and reports material to law enforcement with assistance from the National Center for Missing & Exploited Children, in line with legal requirements. Behavioural signals and heuristic rules are used as well, allowing the company to flag suspicious activity even when individual posts do not appear overtly abusive.
The scale of the problem remains vast. Meta says that between January and June 2026 it actioned 33.2 million pieces of child sexual exploitation content globally across Facebook and Instagram, with more than 97% detected before users reported it. In India, it says it actioned 5.3 million pieces during the same period, with more than 98% proactively identified. The company has also expanded co-operation with Indian authorities by reporting child-safety cases directly to the National Cyber Crime Reporting Portal, operated by the Indian Cybercrime Coordination Centre, or I4C. Separate reporting has shown regulators intensifying scrutiny of Meta’s child-safety systems, including its advertising controls and the role of AI in moderation, while the company has continued to position platform safety as a mix of automation, human review and law-enforcement collaboration.
Meta’s broader strategy suggests that child-safety enforcement is becoming less about removing single posts and more about disrupting the infrastructure around them. That includes advertisement flows, link destinations, repeat offenders, account relationships and the wider network of signals that can reveal a coordinated operation. The same logic is also visible in Meta’s recent anti-scam measures and its efforts against AI-generated explicit content, indicating a wider shift towards using machine learning to identify harmful behaviour before it spreads. The company’s own figures point to the scale of that task, but they also underline a crucial limit: automated systems can improve detection, yet they cannot by themselves eliminate a problem that is constantly evolving.
Disclaimer: This content is intended for informational purposes only. Readers are advised to exercise their own judgement, conduct due diligence, or consult a qualified expert before acting on any information provided.
