Google upgrades federated learning with TEEs for verifiable differential privacy, live in Gboard
ymatias · x · 2026-10-03
Google Research unveiled the next generation of federated learning using Trusted Execution Environments (TEEs), enabling externally verifiable differential privacy while moving heavy computation server-side. Already live in Gboard for English and Japanese next-word prediction, it delivers higher accuracy and substantially faster training cycles. The post credits Katharine Daly, Daniel Ramage and teams, linking the blog and paper.
More from Safety
- AI governance scholar warns the superintelligence race may be only ~8 months from its target — LuizaJarovsky · 2026-10-11
- UK government backs all 44 recommendations of NHS doctors-led AI in healthcare commission — emmanuelvivier · 2026-10-11
- Anthropic documents unexpected model behaviors: real form submissions and access-limit bypass attempts — emmanuelvivier · 2026-10-11
- Beijing reportedly weighing limits on foreign access to China's frontier AI models — emmanuelvivier · 2026-10-11
- Goodfire launches 'inside-out' monitors that catch rogue AI agents at a fraction of the cost — emmanuelvivier · 2026-10-11
- Ben Todd: the "pacing is just regulatory capture" argument makes no sense — ben_j_todd · 2026-10-11