Brief IA

Anthropic: Claude Now Controls Your MacOS

💻 Code & Dev·Tom Levy·

Anthropic: Claude Now Controls Your MacOS

Anthropic: Claude Now Controls Your MacOS
Key Takeaways
1Anthropic has introduced a feature allowing Claude to control your MacOS for various tasks, such as sending files.
2Based on OpenClaw, this innovation enables the AI to execute commands autonomously, with enhanced security measures.
3The combination with Dispatch allows Claude to perform tasks like creating a morning briefing or running tests.
💡Why it mattersThis technological advancement could transform the interaction with our devices while posing crucial security challenges.
Le brief IA que lisent les pros

Le brief IA que les pros lisent chaque soir

Les 7 actus IA du jour, décryptées en 5 min. Gratuit.

Inclus dès l'inscription : notre sélection des meilleurs guides & comparatifs IA.

Choisis ton rythme

Gratuit · Pas de spam · Désabonnement en 1 clic

📄
Full Analysis

Anthropic recently announced a significant advancement for its AI model, Claude, which can now take control of your MacOS computer to perform various tasks. Among these tasks is the ability to send a file left on your hard drive. This feature is accessible to users subscribed to specific plans, thus offering a new dimension to the use of AI.

This innovation is part of the ongoing rise of the open-source framework OpenClaw, which has seen growing popularity earlier this year. OpenClaw has enabled the development of an ecosystem of AI tools, called "claws," capable of executing simple commands in a semi-autonomous manner. Nvidia recently launched NemoClaw, a framework that facilitates the setup and installation of OpenClaw while integrating security options.

Claude uses connectors with applications like Google Calendar or Slack to search for the necessary tools to accomplish a task. If these connectors are not available, Claude can execute tasks manually by simulating the use of a keyboard and mouse. This capability includes the use of programs such as web browsers, development tools, and open files.

When performing these tasks, Claude manipulates the computer like a human user, scrolling and clicking, but always asks for permission before proceeding. Users have the option to interrupt its actions at any time. However, this feature raises security concerns. Experts warn that agentic AI, like Claude, can make quick and potentially dangerous decisions. The "claws" could be hijacked by malicious actors, exploiting your personal data.

To mitigate these risks, Anthropic has implemented security measures, including an automatic scanning system to detect vulnerabilities such as command injections. Despite these precautions, Anthropic advises users not to use this feature with applications handling sensitive data, some of which are disabled by default. The feature is new and may contain bugs, prompting caution.

This feature is currently available for subscribers of Claude Pro and Claude Max, but only on MacOS. It works in tandem with Dispatch, another feature that allows users to assign tasks to Claude via a phone. This includes actions like checking emails or opening Claude Cowork or Claude Code sessions.

The combination of computer use for Claude and Dispatch allows for further task automation, even in the absence of the user. For example, it can create a morning briefing or run tests. Anthropic acknowledges that these tools are still in development and has released this overview to gather feedback and improve Claude's efficiency.

Brief IA — L'actualité IA en français

L'essentiel de l'actualité de l'intelligence artificielle, décrypté et expliqué chaque jour.