Advances in Trust and Security in Wireless Cellular Networks in the Age of AI

Executive Summary

As technology continues to evolve, cellular wireless network organizations face the dual challenge of integrating new innovations while ensuring they remain trustworthy and secure. At the same time, the threat landscape is expanding, with malicious actors exploiting these advancements for harmful purposes. This paper provides a high-level overview of the upcoming security and trust challenges posed by Artificial Intelligence (AI).

Adoption of AI/ML based solutions has gained and continues to gain traction for diverse use cases and mobile networks are no exception. The majority of mobile network related AI/ML solutions to date can be considered proprietary, such as anomaly detection, performance improvements, and increased automation. However, studies to identify appropriate solutions to standardize aspects of AI/ML lifecycle management are currently ongoing. For example, 3GPP has already standardized one solution for AI/ML model training and inference in the 5G Core Network but ongoing studies continue. The O-RAN Alliance has also developed specifications related to AI/ML in the RIC (RAN Intelligent Controller) and has published a report on AI/ML security. Other international standardization bodies, such as ISO, have also been publishing standards to address risks specific to AI, both from organizational and product perspectives. For an in depth understanding of current and planned usage of AI in Cellular Networks readers are encouraged to read the 5G Americas white paper titled “AI for Cellular Networks”.

AI/ML solutions, including those used to protect networks, present an additional attack surface that an adversary can potentially target. On the other hand, an adversary could potentially use AI/ML as an attack vector to launch an attack on a network. It is therefore imperative that AI/ML assets used in mobile networks, such as training/test/validation data and trained models, and their associated parameters/hyperparameters, are protected from unauthorized access, tampering and theft. Equally important is that platforms, where AI/ML Assets are stored and/or processed, are secured in a robust manner. Currently available security best practices and frameworks can be leveraged with strong attention paid to securing AI/ML Assets and platforms.

Also increasing the threat landscape is the implementation of AI platforms. Although AI can potentially be used as a trust and security tool, AI must be designed, developed, and deployed in a secure way.

AI/ML platforms are used across wireless networks for a variety of tasking. Advances and investment in AI will increase its commonality in our wireless networks, functions, and tasking. However, the introduction of an AI platform comes at an increased risk. AI platforms are an attack vector and can also suffer from design and implementation failures. Utilization of secure by design practices and AI threat mapping and risk assessments are essential to ensure products and solutions are secure and risks are mitigated.

AI platforms, models, and/or data are often acquired from third party sources or vendors. Even when deploying these, it is essential organizations seek to thoroughly understand the risk and implement mitigations to ensure AI solutions are designed secure and maintained responsibility.

AI can also be used in conjunction with traditional wireless network attacks, such as eavesdropping, jamming, and spoofing. There are also new, and advanced attacks leverage AI on networks. We recommend investing not only in AI technologies, but in AI security solutions, to maintain pace with the changes threat landscape.

Regulations, frameworks, and standards are emerging globally to aid organizations in developing responsible and trustworthy AI. In the U.S., the Government and NIST also play an important role in aiding in the trust and security of AI platforms.