---
type: "Concept"
title: "Membership inference & gradient leakage"
description: "Privacy attacks that recover information from model updates."
tags: [fl, privacy, threat, federated]
generated: { by: human:crpage, at: 2026-07-09T09:44:00Z }
status: stable
---

Attacks that determine whether a record was in the training set, or reconstruct data from gradients — evidence that FL is **not inherently private enough** for regulated public services. Motivates [secure aggregation](secure-aggregation.md) and [differential privacy](differential-privacy.md).