Computer-use tools are functions that let a model operate a graphical interface the way a person would: reading screens, moving a cursor, clicking buttons, typing text, and submitting web forms. They extend a language model from conversation into action, allowing it to use browsers, desktop apps, and remote services.
In cyber security, these tools matter because they turn prompt-driven software into a capable operator. That is useful for legitimate automation, but dangerous if the model is given too much trust or access. An attacker may abuse them to complete phishing flows, create accounts, trigger transactions, or explore weak web applications at machine speed. Defenders should treat them like privileged automation: limit permissions, isolate sessions, log actions, and require approval for sensitive steps. Strong access control and session management reduce the risk that a tool-enabled model can be misused through ordinary web flaws.