Meta will listen into AI conversations to personalize ads (www.theregister.com)

Meta announced it will begin harvesting users’ text and voice interactions with its Meta AI products — including the web-based chat and integrations in Facebook, Instagram, WhatsApp and Messenger — to generate personalized posts, reels and advertising starting December 16, 2025, with user notifications beginning October 7, 2025. There is no global opt-out and the change will not initially apply to users in the EU, UK or South Korea. Meta says it will not use conversations that touch on religion, sexual orientation, politics, health, race/ethnicity, philosophical belief or trade-union membership, but that carve-out also creates easy workarounds and enforcement questions. For the AI/ML community this is significant because it turns conversational data (text and voice) into a closed-loop signal stream for personalization and ad-targeting models — effectively a new, rich training and attribution source tightly coupled to Meta’s ad stack. That promises stronger engagement and automated creative/targeting workflows but raises technical and ethical issues: opaque attribution, reduced external auditability of ad performance, higher risk of sensitive-inference or model drift despite stated exclusions, and major privacy/regulatory implications. Practitioners should expect demands for stronger dataset governance, auditing tools, provenance tracking and privacy-preserving techniques if such conversational signals become a mainstream input for recommender and ad models.