Privacy and Trust in Peer-to-Peer Federated Learning
Removing the central server from federated learning removes a single point of failure — and a single point of trust. Notes on what replaces it.
Bassam Alotaibi
AI Governance & Cybersecurity Researcher
Classical federated learning keeps data local but still depends on a central coordinator to aggregate model updates. That coordinator is a privacy chokepoint, an availability risk, and an implicit trust anchor. Peer-to-peer topologies remove it — and in doing so, redistribute every problem it was quietly solving.
Without a centre, each participant must decide for itself which peers to learn from, how much to reveal in each exchange, and how to detect a neighbour that is malicious, compromised, or simply wrong. Static privacy budgets and fixed trust assumptions handle none of this well: the environment is dynamic, and the protection should be too.
My current research direction explores adaptive mechanisms — privacy that tightens when trust signals weaken, trust models that update from observed behaviour rather than declared identity, and secure exchange protocols that make honest participation the rational strategy. The aim is collaborative learning that degrades gracefully under attack instead of failing absolutely.