AI in cybersecurity: The good, the bad, and the FUD

Executive summary: The report describes how adversaries—including nation-state actors—are leveraging large language models and Model Context Protocol (MCP) servers to automate reconnaissance, phishing, and other tradecraft, and highlights primary risks to AI infrastructure such as prompt-injection/model hijacking and supply-chain exposure. It emphasizes returning to core information-security controls (least privilege, defense-in-depth, credential management, environment segmentation) and prescribes operational guidance for safely adopting AI agents in SOC workflows while maintaining human oversight.