Meta adds AI tools to its ad review to fight child exploitation

Ads as a new route for predators

According to Meta, predators move across platforms and change tactics to avoid detection, and advertising is one of the surfaces where it has seen this shift. The company says bad actors often use content designed to evade review. An ad can look harmless on its own while directing people to illegal material hosted off Meta's platforms.

Once Meta understood how this malicious advertising scheme worked, it says it did not limit its response to the ads that had been reported. It widened its investigation to find related activity across its platforms, disabled the violating accounts behind it and blocked the off-platform links involved.

What is new in ad review

Meta says it then deployed additional measures, including new AI tools, to improve how its systems identify ads that either feature CSAM or send people to violating activity off-platform. The list in the announcement includes:

• LLM detection of "signposting": a new large language model system aimed at seemingly benign ad content that is strongly suspected of covertly directing people to illegal content or harmful activity off Meta's platforms.
• Destination analysis: an improved understanding of where an ad leads, not only what it shows, so Meta can block violating destinations and act against the accounts responsible.
• Additional AI-driven sweeps: checks of ad content to surface material that earlier systems did not catch, using signals the company says it will keep expanding as its investigation continues.
• A red-teaming AI agent: a tool that probes Meta's own defences to find weaknesses, so new adversarial tactics can be identified before they scale.
• Stronger recidivism detection: mechanisms for spotting bad actors who try to create new accounts after being removed.

Existing detection technology and the numbers

Meta also outlined the tools it already uses. These include behavioural signals that combine metadata and activity patterns to find accounts acting suspiciously, and heuristics-based rules that combine classifiers, AI models and other signals to catch new violation patterns at scale, including in ads.

The company says it has used PhotoDNA and other photo and video-matching technology across all of its apps since 2011 to detect identical or near-identical copies of known CSAM. According to Meta, it maintains one of the industry's largest databases of image and video hashes, adds newly identified hashes to it and shares them through the Tech Coalition's Lantern program so other participating companies can remove the same content. The same technology is used to identify CSAM in off-platform link previews.

Meta says it blocks links to external websites that host or create violating CSAM. After a link is blocked, it searches for and deletes other content containing it, such as ads, posts or comments. People are prevented from posting blocked links on Facebook, Instagram and Threads, and ads containing them are rejected at upload.

On scale, Meta reports that between January and June 2026 it actioned 33.2 million pieces of child sexual exploitation content from Facebook and Instagram globally, with over 97% found and proactively addressed before anyone reported it. In India, it actioned 5.3 million pieces over the same period, with over 98% found proactively.

Law enforcement reporting and what it means

Meta says that when it finds potential child sexual abuse or exploitative content, it investigates, removes it if it violates policy, disables accounts where signals point to malicious sharing, and reports to the National Center for Missing & Exploited Children (NCMEC) in line with applicable law. It adds that a team including former FBI investigators runs manual investigations to uncover networks of predatory accounts, with the aim of acting at network level rather than account by account.

The company also said that in September it announced it will report child safety cases directly to India's National Cyber Crime Reporting Portal (cybercrime.gov.in), run by the Indian Cybercrime Coordination Centre (I4C). This builds on its existing reporting of child sexual abuse material through the portal with NCMEC's assistance.

For users and advertisers, the changes mean closer scrutiny of where an ad leads, not only what it displays, and a higher chance that accounts tied to violating ads are disabled and their links blocked. Meta presents the work as ongoing, saying it will keep adapting its tools as offenders change their methods.