Abstract

The growing reliance on artificial intelligence (AI) for cybersecurity applications, particularly in threat intelligence sharing, presents both significant opportunities and complex challenges. As cyber threats become more sophisticated and widespread, organizations increasingly depend on collaborative efforts to detect, analyze, and mitigate attacks. Sharing threat intelligence across sectors enhances situational awareness and strengthens collective defense mechanisms. However, this collaborative approach raises serious privacy concerns, as the data involved often contains sensitive, proprietary, or personally identifiable information (PII). Ensuring data confidentiality while enabling meaningful AI-driven analysis necessitates robust privacy-preserving techniques. Homomorphic encryption (HE) has emerged as a promising solution that allows computations to be performed directly on encrypted data, preserving privacy without compromising analytical utility. This capability enables organizations to share encrypted threat intelligence and perform AI-based anomaly detection, malware classification, and pattern recognition without revealing raw data. In this context, HE facilitates secure multi-organizational collaboration, allowing encrypted inputs to be processed by machine learning models without exposing underlying information. This explores the integration of homomorphic encryption with AI in cybersecurity, emphasizing its application in privacy-preserving threat intelligence sharing. We review the various types of homomorphic encryption partial, somewhat, and fully homomorphic encryption and evaluate their suitability for real-world cybersecurity use cases. System architectures that support encrypted AI model inference, encrypted feature extraction, and secure aggregation of threat data are examined. Additionally, we discuss trade-offs between encryption strength, model performance, latency, and computational overhead. Case studies from sectors such as finance and healthcare illustrate the practical feasibility of HE-enhanced cybersecurity frameworks. Ultimately, this highlights homomorphic encryption as a key enabler of privacy-preserving AI for cybersecurity. It underscores the importance of continued research and optimization to make HE-based threat intelligence systems scalable, efficient, and widely adoptable in the evolving landscape of cyber defense.