Digital Economy

Claude takes control of the PC (but asks permission): Anthropic's AI becomes an operating assistant

With the Dispatch function, Claude can execute commands on the computer directly from his smartphone, create files, manage apps and automate tasks

by Riccardo Saporiti

 Alamy Stock Photo

2' min read

Translated by AI
Versione italiana

2' min read

Translated by AI
Versione italiana

The point, of course, is to leave the computer accessible. The advantage, or at least this is what they are convinced of at Anthropic, is that it will be possible to give Claude commands to execute on the machine, whether portable or stationary, directly from your smartphone.

On the desktop side, you need to have installed the app for this AI model, which provides the options Claude Cowork and Claude Code, the latter of which specialises in programming, which have the ability to create and edit files on the PC or, better still, on a specific folder to which they have been given access. Both on this and on your phone, you need to enable the Dispatch function and that's it.

Loading...

Are you running late for a business appointment? You can ask Claude to save the keynote you have to present and attach it to the appointment on your calendar so that it is available to all meeting participants. Are there photos to be cropped in a folder? It is possible to put the Anthropic AI to work while the human counterpart is busy with other tasks.

To understand which software to use, Claude will start with the ones he is authorised to use. It is indeed possible to connect this model to programmes such as Slack or Google Calendar. A solution that, by exploiting the API, would reduce the margins of error. Should he not find them, he would take control of the mouse, keyboard and screen to complete the assigned task.

Now, giving this kind of access to one's computer to a machine rings more than a few alarm bells. They are also aware of this at Anthropic and in fact, in launching this functionality, the company has specified that before taking control of the PC and carrying out any action, it will ask the human user for permission.

Not only that, there are mechanisms in place to reduce risks, starting with prompt injection monitoring. This is a cyber attack that attempts to trick an LLM by means of instructions hidden in the text of a website that invite the AI to download a virus. And then there is the off button: the human user can stop Claude at any time

Despite these safeguards, however, it is Anthropic itself that is putting its hands up. In the post announcing the introduction of this function, it explains that 'it is still in its infancy compared to Claude's ability to write code'. The model 'can make mistakes, and while we continue to improve our protection mechanisms, so do the threats'. That is why 'we recommend only starting with applications that you trust and not working with sensitive data'.

Copyright reserved ©
Loading...

Brand connect

Loading...

Newsletter

Notizie e approfondimenti sugli avvenimenti politici, economici e finanziari.

Iscriviti