AI Agent (Automated Browsing)

An AI browsing agent pairs a large language model with a controllable browser, usually a headless or instrumented one. The model receives the current page state, whether as rendered content, structured accessibility data, or screenshots, reasons about the goal it was given, and issues actions such as clicking, typing, and navigating. It then observes the result and decides its next step, iterating until the task is done.

Unlike a traditional bot that replays a hardcoded sequence, an agent adapts to layouts it has not seen before and recovers from unexpected states. This flexibility lets it complete open-ended tasks like researching, filling forms, comparing options, or making purchases, and it means the agent's behavior is less repetitive and therefore harder to catch with simple pattern rules.

Underneath, however, an agent still runs inside an automated browser and reaches sites over some network path. It inherits the same environmental and control-channel signals as other automation: headless or instrumented browser characteristics, automation-framework artifacts, and the origin and reputation of its connection. Its reasoning is novel, but its execution substrate is familiar.

The rise of agents also creates a legitimacy question rather than a purely adversarial one. Some agents act on behalf of a real user with permission, while others are used for abuse. This pushes detection toward identifying that a session is agent-driven and providing that context, so each service can apply its own policy rather than assuming all agent traffic is malicious.