FedSecureFormer Introduces Lightweight Federated Transformer for Vehicle Intrusion Detection

Global: FedSecureFormer: Fast Federated Transformer for Intrusion Detection in Connected Vehicles

Researchers led by Devika S and colleagues released a preprint on December 30, 2025 describing FedSecureFormer, a lightweight, encoder‑only transformer designed to detect intrusions in connected and autonomous vehicles while operating within a federated learning framework.

Authors and Publication Details

The paper lists five contributors—Devika S, Vishnu Hari, Pratik Narang, Tejasvi Alladi, and F. Richard Yu—and was submitted to arXiv under the categories Cryptography and Security (cs.CR) and Artificial Intelligence (cs.AI). The work is identified by arXiv ID 2512.24345 and carries a DOI of 10.48550/arXiv.2512.24345.

Model Architecture

FedSecureFormer employs an encoder‑only transformer with a deliberately reduced number of layers, aiming to minimize computational overhead without sacrificing detection accuracy. By focusing on essential self‑attention mechanisms, the model can be deployed on the limited hardware typical of vehicle‑edge nodes.

Federated Learning Integration

The framework leverages federated learning to train the transformer across multiple vehicles or edge devices. Model updates are aggregated centrally, allowing the system to improve its detection capabilities while raw sensor data remains on‑device, thereby enhancing privacy and reducing bandwidth requirements.

Potential Impact on Vehicle Security

If adopted, the approach could provide real‑time threat identification for connected and autonomous vehicles, addressing a growing concern as automotive systems become increasingly software‑driven. The authors argue that the combination of a compact model and decentralized training aligns with the latency and security constraints of vehicular networks.

Next Steps and Availability

The authors indicate that further empirical evaluation on benchmark intrusion datasets is planned, and they intend to release code and model weights alongside the preprint. Interested parties can access the full paper through the arXiv platform.

This report is based on information from arXiv, licensed under Academic Preprint / Open Access. Based on the abstract of the research paper. Full text available via ArXiv.