Definition
A way of training an AI model across many separate devices or organizations' data without the raw data ever leaving its original location — only the learned updates are shared and combined centrally. It's often used for sensitive data such as hospital records or personal devices, where moving raw data around is undesirable or restricted.
Why it matters
It's often marketed as privacy-friendly since data stays local, but the software coordinating the shared updates is itself a high-value target — a flaw there can lead to remote code execution across every participating organization at once.