Communication method and apparatus for AI network security

ChangeBridge: Patent Grants - Networking (H04L)

Published March 31st, 2026

Detected March 31st, 2026

Summary

USPTO granted Patent US12592960B2 to Huawei Technologies Co., Ltd. on March 31, 2026, for a communication method improving AI network security via anti-attack policies. The patent covers obtaining anti-attack policy information, determining training or inference data based on that information, and using data source blocklists, trustlists, preprocessing methods, or data enhancement techniques to protect AI models in network environments.

View original document View source feed page

What changed

USPTO granted Patent US12592960B2 to Huawei Technologies Co., Ltd. on March 31, 2026, for a method and apparatus related to AI network security via anti-attack policies. The patent discloses obtaining first information indicating an anti-attack policy of first data (which includes training data and/or inference data of a model), determining the first data based on that information, and sending the first data for model training or inference. The anti-attack policy can include data source blocklist information, data source trustlist information, data preprocessing manner information, or data enhancement manner information.

This is a patent grant notice and does not impose compliance obligations on third parties. Technology companies developing AI systems in network environments may wish to review this patent when designing security mechanisms to avoid potential infringement claims. No immediate action is required unless the company is actively designing similar anti-attack policy mechanisms for AI model security.

Source document (simplified)

← USPTO Patent Grants

Communication method and apparatus

Grant US12592960B2 Kind: B2 Mar 31, 2026

Assignee

HUAWEI TECHNOLOGIES CO., LTD.

Inventors

Yu Zeng, Tingting Geng, Qinghai Zeng

Abstract

The present disclosure provides a communication method and apparatus, to improve security of applying artificial intelligence in a network. The method includes obtaining first information, where the first information indicates an anti-attack policy of first data, and the first data includes training data and/or inference data of a model; determining the first data based on the first information; and sending the first data, wherein the first data is used to perform model training and/or determine an inference result. The first information can be at least one of the following: data source blocklist information of the first data and/or data source trustlist information of the first data, information indicating a data preprocessing manner of the first data, or information indicating a data enhancement manner of the first data.