Meta added new AI tools to detect ads covertly directing users to child exploitation material. It now uses large language models to flag signposting content. The system also blocks off-platform links leading to illegal material. A red-teaming AI agent probes defenses for new adversarial tactics. Recidivism detection identifies bad actors creating new accounts after removal. Between January and June 2026 Meta actioned 33.2 million pieces of such content globally. Over 97% were found proactively before any user report. In India 5.3 million pieces were actioned in the same period. Over 98% of those were found proactively. Meta reports cases directly to India’s National Cyber Crime Reporting Portal. It also shares hashes with industry partners via the Tech Coalition’s Lantern program.
What changed
Previous detection relied on user reports and known hash databases for CSAM.
What it unlocks
Proactively detecting and blocking covert ads directing users to off-platform child exploitation material.
- 97% found proactively before reporting
- 5.3 million pieces in India Jan-Jun 2026
- 98% found proactively in India
Sources