AI in cybersecurity – how do we handle cybersecurity reports and triage them with LLMs

Abstract

The integration of large language models (LLMs) into cybersecurity operations represents a significant shift in how threats are detected, categorized, and mitigated. This paper aims to explore the application of LLMs in automating the triage process of cybersecurity incident reports, with a focus on real-world implementation within Hackout.ro, a Romanian civic platform designed to crowdsource and process user-submitted reports of phishing, fake news, AI-generated content, and other digital threats. Building on prior work in AI-assisted threat intelligence and natural language processing for security applications, this study positions LLMs not only as tools for efficiency but also as agents of digital empowerment for non-technical users. Using a case study approach, we analyze workflows where user-submitted content is automatically classified, prioritized, and, in certain cases, responded to in real time. Additional tools enable non-expert users to test and secure their applications using AI-assisted guidance. Our findings demonstrate that LLMs can significantly reduce response time, enhance classification accuracy, and facilitate proactive defense strategies. However, the dual-use nature of AI is also evident, as malicious actors increasingly exploit the same technologies for sophisticated attacks.The implications of this research are far-reaching: security practitioners gain new tools for automation, educators can incorporate AI literacy into digital resilience programs, and policymakers must rethink regulatory frameworks for AI in cybersecurity. The key contribution of this paper is a detailed, practice-based insight into how LLMs are currently operationalized in cyber defense workflows, highlighting both opportunities and limitations. By presenting a balanced and empirically grounded perspective, the study supports a more ethical, efficient, and inclusive approach to cybersecurity in the age of artificial intelligence.