The application of this artificial intelligence model must be installed on the desktop side. Thanks to the Claude Cowork and Claude Code options within the application, which focus specifically on software development, it is possible to create and edit files in certain folders that users allow. The system is ready by activating the “Dispatch” feature on both the computer and the phone.
For example, if you are late for a business meeting, Claude can be asked to save the presentation file and add it to the meeting in the calendar. Or, if photos in a folder need to be cut, it is possible to delegate this task to artificial intelligence while the user deals with other tasks.
When determining which software to use, Claude first starts with the applications to which access is granted. It can be connected to programs such as Slack or Google Calendar. Since these integrations work via APIs, they reduce the margin of error. If it cannot find the necessary tools, it can directly take control of the mouse, keyboard and screen to complete the given task.
Of course, this type of access raises security questions. Anthropic is also aware of this. The company emphasizes that when activating this feature, Claude will definitely ask for permission from the user before performing any action.
However, various security mechanisms have also been developed to reduce risks. One of these is the monitoring system against cyber attacks known as “prompt injection”, which aims to deceive artificial intelligence with harmful commands. The user also has a “turn off” option where he can stop Claude whenever he wants.
However, Anthropic states that caution should be exercised. The company states that this new feature is still in its early stages compared to Claude’s coding abilities. Reminding that the model may make mistakes and threats are constantly evolving, users are advised to start only with applications they trust and avoid working with sensitive data.
