FedSpy-LLM: Towards Scalable and Generalizable Data Reconstruction Attacks from Gradients on LLMs

2026-04-07

AI Summary

Existing gradient inversion attacks struggle to achieve high-fidelity, scalable, and architecture-agnostic reconstruction of training data from large language models (LLMs) in federated learning settings employing parameter-efficient fine-tuning (PEFT). This work proposes FedSpy-LLM, which introduces a novel gradient decomposition strategy to exploit the rank deficiency and subspace structure inherent in gradients, enabling efficient extraction of salient tokens. Coupled with an iterative alignment mechanism to recover sequence ordering, FedSpy-LLM achieves, for the first time, high-quality reconstruction of batched, long-sequence training data from PEFT-adapted LLMs. The method supports diverse architectures—including encoder-only, decoder-only, and encoder-decoder models—and significantly outperforms existing approaches in both reconstruction fidelity and scalability, thereby exposing critical privacy vulnerabilities in current real-world deployments.

Abstract

Given the growing reliance on private data in training Large Language Models (LLMs), Federated Learning (FL) combined with Parameter-Efficient Fine-Tuning (PEFT) has garnered significant attention for enhancing privacy and efficiency. Despite FL's privacy benefits, prior studies have shown that private data can still be extracted from shared gradients. However, these studies, mainly on full-parameter model training, are limited to reconstructing small batches, short input sequences, and specific model architectures, such as encoder-based or decoder-based models. The reconstruction quality becomes even worse when dealing with gradients from PEFT methods. To fully understand the practical attack surface of federated LLMs, this paper proposes FedSpy-LLM, a scalable and generalizable data reconstruction attack designed to reconstruct training data with larger batch sizes and longer sequences while generalizing across diverse model architectures, even when PEFT methods are deployed for training. At the core of FedSpy-LLM is a novel gradient decomposition strategy that exploits the rank deficiency and subspace structure of gradients, enabling efficient token extraction while preserving key signal components at scale. This approach further mitigates the reconstruction challenges introduced by PEFT's substantial null space, ensuring robustness across encoder-based, decoder-based, and encoder-decoder model architectures. Additionally, by iteratively aligning each token's partial-sequence gradient with the full-sequence gradient, FedSpy-LLM ensures accurate token ordering in reconstructed sequences.

Problem

• data reconstruction attack
• federated learning
• large language models
• parameter-efficient fine-tuning
• gradient leakage

Innovation

• gradient inversion
• federated learning
• parameter-efficient fine-tuning
• data reconstruction attack
• large language models