Meta actioned 33.2 million pieces of child sexual exploitation content across Facebook and Instagram globally in the first half of 2026 — and now it is adding new AI tools to its ad review process to stop bad actors from directing users towards such material hosted off-platform.

The company says it observed advertisers shifting tactics to evade detection, including ads that look harmless on their own but carry signals covertly steering users towards illegal material elsewhere. After identifying the pattern, Meta widened its investigation across its platforms, disabled violating accounts and blocked associated links.

What exactly has shipped?

The update package includes several concrete additions:

• LLM-based detection for "signposting" — seemingly benign ad content suspected of pointing users towards child sexual exploitation material or other harmful activity outside Meta's platforms.
• Destination-based review — systems now evaluate where an ad ultimately sends users, not just what appears inside the ad, so violating destinations and linked accounts can be actioned.
• Additional AI-driven sweeps of advertising content to surface material older systems may have missed.
• A red-teaming AI agent that probes Meta's own detection systems for weaknesses and surfaces emerging evasion tactics.
• Stronger detection of previously removed users attempting to open new accounts.

Beyond advertising, Meta combines behavioural signals, AI models and photo/video-matching technologies, including PhotoDNA, to identify suspected CSAM. It shares newly identified image and video hashes with participating companies through the Tech Coalition's Lantern programme.

How does link enforcement work?

When Meta blocks a link to an external website tied to CSAM, it doesn't stop there. The company searches for other ads, posts and comments carrying the same link, prevents the blocked link from being posted on Facebook, Instagram and Threads, and rejects ads containing it at upload.

Meta also runs investigations into networks of accounts suspected of predatory activity — a team that includes former FBI investigators. Content flagged internally or reported externally is reviewed, removed where it violates policy, and accounts face disablement where malicious sharing is identified.

The enforcement volume is large, and mostly proactive. More than 97% of the 33.2 million pieces actioned globally between January and June 2026 were caught before users reported them. In India, Meta actioned 5.3 million pieces over the same period, with more than 98% identified pre-report.

Why is Meta under pressure on this?

The measures follow sustained scrutiny of how child sexual abuse material slips through Meta's ad systems. An investigation by the Tech Transparency Project (TTP) found Meta ran 332 paid ads containing CSAM on Facebook and Instagram this year. Most allegedly used AI-manipulated images of children to depict sexual activity; TTP also identified images of real children lifted from social media posts and websites and used in the ads.

Responding at the time, a Meta spokesperson said the company "does not tolerate child exploitation, whether involving real or AI-generated material". The spokesperson added that many of the flagged ads had already been removed, most received fewer than 200 impressions, and total ad spend came in below US$5,000 — while pointing to the changing tactics used by those trying to evade detection.

India has also ordered Meta to take down child sexual abuse ads, adding regulatory pressure on top of watchdog findings.

For advertisers and agencies, the practical takeaway is straightforward: ad review on Facebook and Instagram now examines destination URLs as well as creative, and repeat offenders face harder account-recreation paths. Campaigns with innocuous-looking creative pointing to flagged external destinations will be rejected at upload.