Agentic AI: Balancing Benefits and Risks in Cybersecurity

In a recent interview with Security Magazine, Diana Kelley, Chief Information Security Officer (CISO) at Noma Security, discussed the advantages, risks, and best practices associated with agentic AI. Agentic AI refers to autonomous or semi-autonomous AI systems capable of performing tasks, making decisions, and interacting with minimal human intervention. Kelley highlights the potential benefits of agentic AI, such as task automation and improved efficiency, which can significantly enhance operational workflows. However, these benefits come with notable risks, particularly in security and privacy. Autonomous systems can introduce new vulnerabilities, especially if not properly secured or monitored. Kelley emphasizes the importance of robust risk management and continuous monitoring to mitigate these risks. For cybersecurity professionals, the adoption of agentic AI requires careful planning to ensure that these systems do not become liabilities. The impact on the cybersecurity landscape is substantial, as agentic AI can both enhance threat detection and response while also creating new attack surfaces. Organizations must implement stringent governance frameworks, regular audits, and adaptive security controls to manage these systems effectively. Kelley’s insights underscore the need for a balanced approach, leveraging the efficiency gains of agentic AI while maintaining rigorous oversight to prevent misuse or unintended consequences. Agentic AI systems, by their nature, operate with a degree of independence that necessitates robust security measures. For instance, in cybersecurity operations, an autonomous threat detection system might analyze network traffic, identify anomalies, and respond to incidents without human intervention. While this can drastically reduce response times and mitigate threats more efficiently, it also poses risks if the AI’s decision-making processes are compromised. Adversaries could exploit vulnerabilities in the AI’s algorithms or data inputs to manipulate its actions, leading to false positives, overlooked threats, or even malicious actions. Kelley’s emphasis on continuous monitoring aligns with the dynamic nature of AI systems. Unlike static software, agentic AI can evolve through machine learning, meaning its behavior may change over time. This necessitates ongoing oversight to ensure that the AI remains aligned with organizational security policies and does not deviate into unsafe or unpredictable actions. Additionally, risk management must be proactive, anticipating potential failure modes and ensuring that safeguards are in place to limit damage if something goes wrong. From a practical standpoint, cybersecurity teams should consider implementing agentic AI in phases, starting with non-critical tasks to assess performance and security implications before scaling up. They should also invest in explainable AI (XAI) techniques to ensure transparency in decision-making processes, which is crucial for accountability and debugging. Furthermore, integrating AI systems with existing security infrastructure, such as SIEM (Security Information and Event Management) platforms, can help maintain visibility and control. In conclusion, while agentic AI offers transformative potential for cybersecurity, its adoption must be approached with caution. Organizations should prioritize security-by-design principles, ensuring that AI systems are built with robust protections from the outset. Regular audits, continuous monitoring, and a strong governance framework are essential to harness the benefits of agentic AI while minimizing its risks.