Abstract The paper explores the multifaceted impacts of the integration of the increasingly autonomous AI agents into the field of cybersecurity. The paper attempts to address the technical, business, ethical, and legal considerations associated with the increasing autonomy of security tools and solutions. The discussion starts by shedding light on the legal qualification of AI agents, and the paper proposes to distinguish between automated AI and AI agents. The basis of this distinction is legal, especially regarding liability in case of fully autonomous AI agent. A systematic review protocol in accordance with PRISMA guidelines was undertaken to elaborate a comprehensive analysis of the effects of increasing autonomy in AI security agents. Key technical limitations were identified, business considerations were discussed, and ethical and legal challenges were analyzed. Despite awaiting various challenges, literature suggests that full autonomy of AI agents is becoming progressively feasible and offers a competitive advantage to organizations in specialized cybersecurity fields. Finally, the paper strongly advocates for the study of future research gaps to ensure that the integration of agentic AI systems into the field of cybersecurity leads to robust, safe, and effective security solutions.
El-Khoury et al. (Tue,) studied this question.