Page 86 - Cyber Defense eMagazine September 2025

Applying Agentic AI in Security

Penetration Testing

Agentic AI can significantly enhance penetration testing (pentesting) by automating the initial stages of
vulnerability discovery. AI agents can rapidly scan attack surfaces, and identify common security flaws.
This enables human pentesters to focus on more sophisticated attack vectors, leading to deeper and
more comprehensive security assessments.

Threat Detection and Response

Security Operations Center (SOC) teams are often overwhelmed with alerts, many of which turn out to
be false positives. Agentic AI can autonomously investigate these alerts, distinguishing genuine threats
from benign anomalies. By handling routine investigations, agentic AI allows human analysts to
concentrate on high-priority incidents.

Additionally, AI agents can be scaled instantly, effectively doubling a security team’s capacity without
hiring additional staff. This automation enhances efficiency while reducing operational costs.

Vulnerability Management

Prioritizing vulnerabilities has long been a challenge for security teams, requiring extensive context
gathering and analysis. Agentic AI can automate this process by reviewing historical vulnerability data,
assessing potential risks, and recommending remediation strategies. This streamlines the workflow,
ensuring that security teams focus on the most critical threats.

Balancing the Risks and Rewards of Agentic AI

While agentic AI offers significant advantages, it also presents new risks. Organizations must implement
safeguards to prevent unintended consequences and security vulnerabilities.

Guardrails and Human Oversight

Excessive autonomy can be problematic. AI agents must be designed with oversight mechanisms to
prevent misjudgments or unintended actions. Human-in-the-loop workflows ensure that critical decisions
receive human review before execution.

A recent legal ruling involving Air Canada’s AI chatbot highlighted the importance of accountability.
Companies cannot dismiss AI-generated errors as mere software glitches. Instead, they must assume
responsibility for their AI systems and establish clear governance policies.

Mitigating Prompt Injection Attacks

Agentic AI’s ability to gather and process vast amounts of data also introduces security risks. Malicious
actors can manipulate AI agents through prompt injection attacks, steering them toward unintended
actions. This threat underscores the need for robust security protocols and continuous monitoring.

Cyber Defense eMagazine – September 2025 Edition 86
Copyright © 2025, Cyber Defense Magazine. All rights reserved worldwide.