Survey synthesizes dual-use risks and defensive strategies for LLMs in cybersecurity

A comprehensive survey analyzes the dual-use impact of Large Language Models (LLMs) and generative AI on cybersecurity, covering both automated defense mechanisms and sophisticated attack vectors. The authors review over 70 academic papers and industry reports to synthesize insights from platforms such as Google Play Protect, Microsoft Defender, and AWS.

• LLM-generated malware is estimated to account for 50% of detected threats by 2025, a significant increase from 2% in 2021.
• The survey examines beneficial applications including zero-day detection, DevSecOps integration, federated learning, and synthetic content analysis.
• Defensive strategies discussed include explainable AI (XAI), model watermarking, adversarial defense, and the SAFE Framework.
• Practical recommendations emphasize responsible deployment, transparency, and cross-industry collaboration for trustworthy AI systems.

The work aims to provide a roadmap for secure, scalable LLM systems and serves as a critical reference for researchers and security leaders navigating AI-driven threats.