Am I FLoCed? (amifloced.org)

Google has begun a Chrome "origin trial" of Federated Learning of Cohorts (FLoC), an experimental replacement for third‑party cookies that groups users by browsing behavior and exposes a cohort ID to sites and advertisers. The trial affects roughly 0.5% of Chrome users in selected countries (including the U.S., India, Brazil, Japan, etc.) and may expand; Google intends to iterate on grouping algorithms and parameters. FLoC runs in-browser, hashes the list of domains visited in the past seven days using SimHash, assigns each user to one of ~33,000 cohorts (groups of “a few thousand”), and recalculates the cohort weekly. Users can opt out only by disabling third‑party cookies or switching browsers; site owners can opt out of inclusion with the Permissions-Policy: interest-cohort=() header. For the AI/ML community this matters because FLoC changes how behavioral data is produced and shared: it centralizes cohort labels as an API-readable signal that advertisers and platforms can use to train or augment models for ad targeting, profiling, demographic inference, or content personalization. That increases risks of deanonymization, fingerprinting (cross‑site correlation), dataset bias, and disparate impact from automated decision systems. The trial’s opaque deployment and the ease of aggregating cohort-level signals make it both a new data source and a new attack surface—researchers and practitioners should evaluate how cohort signals affect model fairness, privacy guarantees, and the ethics of using browser-derived behavioral labels.