Meta deploys AI to catch ads hiding child abuse links

Meta acted on 33.2 million pieces of child sexual exploitation content on Facebook and Instagram in the first half of 2026, with over 97% detected before users reported it. The company is rolling out new AI tools targeting ads that appear harmless but secretly direct users to illegal content hosted elsewhere. A new large language model system detects this tactic, called signposting, by analyzing where ads send users, not just their content. Meta is also deploying a red-teaming AI agent to find weaknesses in its own safety systems and improving detection of banned users who return with new accounts.