Agentic browsing is web navigation performed by software that can choose its own actions and complete multi-step tasks, such as opening pages, clicking buttons, filling forms, and moving between services. Unlike a simple bot that repeats one scripted action, an agent can decide what to do next based on what it sees in the browser.
This matters in cyber security because an agent may hold real user sessions, credentials, or access to sensitive sites. That turns browsing into an authorization problem: websites must decide whether to allow, throttle, or block automated visitors, and defenders must control what the agent can reach. In real environments, agentic browsing can be abused for account abuse, data collection, or unintended transactions if it is over-privileged or tricked by hostile pages. Good defenses include least privilege, explicit approval for sensitive actions, strong logging, and testing agents against deceptive prompts and unsafe login flows.