On August 14, Google's security blog published How Google is making private AI practical with homomorphic encryption. The claim in the title is the news: homomorphic encryption, long filed under "theoretically beautiful, operationally hopeless," pitched by Google as practical for real AI workloads.
The post drew 473 points and 276 comments on Hacker News inside roughly 28 hours - a level of engagement that applied-cryptography posts rarely get, and a signal that the performance claims are being argued over rather than accepted. Worth reading the primary post and the thread together before taking the "practical" framing at face value; overhead numbers are where this class of announcement usually lives or dies.
Sources: Google Security Blog, HN discussion
