What "Agentic" Means

An agentic AI system is one that can take actions autonomously. It's not just responding to your input — it's executing commands, sending emails, making API calls, browsing the web, running code, interacting with other applications, modifying files. Unlike a conversational AI, which you interact with and then decide what to do with its outputs, an agentic system acts on its own judgment.

This changes the security equation significantly. If a conversational AI tool is compromised by prompt injection, the worst-case outcome is that it produces bad text — text you read, might act on, but at least have the opportunity to verify. If an agentic AI is compromised, it takes bad actions directly. It might send an email without your review. It might delete files. It might transfer funds. It might grant access to systems. The AI itself becomes the attack vector.

This is why agentic AI requires particular attention to security safeguards. The potential impact of a successful attack is higher because the AI isn't just advising you — it's acting.