Microsoft grants Copilot file access and local AI control on Windows PCs
Microsoft announces Copilot will soon manage local files and run on-device AI models for Copilot+ PC users, aiming to lower costs and keep data private.
Key points
- Copilot will gain permission-based access to move files and change settings on Windows PCs.
- The assistant will access local files and recent user activity to provide better context.
- Certain tasks will run on local AI models, accessing the cloud only when necessary.
Microsoft announced plans to expand the capabilities of its Copilot AI assistant on Windows PCs, granting it greater control over local systems. The company stated that these changes will allow the assistant to organize files, modify settings, and execute tasks using on-device artificial intelligence models.
These updates were revealed at Microsoft’s Hybrid Intelligence event, where the company outlined a strategy to blend local and cloud processing. According to Computerworld, the goal is to help customers reduce AI expenses while maintaining tighter control over their data. The new features are scheduled to roll out to Copilot+ PC users in the coming months.
Enhanced Local Capabilities
One of the primary changes involves Copilot’s ability to access local files and recent user activity. This access provides the AI assistant with additional context, improving the relevance of its outputs. Microsoft indicated that this functionality allows the system to understand the user’s current workflow more effectively than before.
Copilot will also be authorized to perform direct actions on the device. These actions include moving files and changing system settings. Jacob Andreou, Microsoft’s Executive Vice President for Copilot, stated that the assistant will have the same level of control as the user, but only with explicit permission.
Hybrid Intelligence Strategy
The company also introduced an option to run Copilot on local AI models for specific tasks. Microsoft said the assistant will access cloud servers only when needed, delegating simpler or privacy-sensitive work to the device. Andreou noted that the cloud will still handle the most complex tasks, but local models will be used when cost or privacy is a priority.
This hybrid approach reflects broader market trends, according to Biswajeet Mahapatra, a principal analyst at Forrester. He noted that enterprises increasingly want the flexibility to choose which workloads run locally and which belong in the cloud, rather than relying exclusively on one method.
Who Is Affected
The new capabilities are specifically targeted at users of Copilot+ PCs. Microsoft stated that these three new features will be available to this group in the coming months. The announcement also included hardware updates, such as the Surface Laptop Ultra and Surface RTX Spark Dev Box, which are designed to support these more efficient AI models.
In a demonstration, Andreou showed how local file access improves Autopilot, a continuous background agent formerly known as Scout. Autopilot, currently in private preview, can also run on locally hosted models. This helps reduce costs and prevents private data from being sent to external cloud servers.
What should you do?
Security professionals should monitor Microsoft’s official channels for detailed documentation on the permission models governing these new Copilot actions. Understanding how the assistant requests and executes changes to local settings is critical for maintaining system integrity.
Organizations should review their data classification policies to determine which files and activities should remain accessible to on-device AI models. Ensuring that sensitive data is excluded from local AI processing contexts can help mitigate privacy risks.
IT administrators should prepare to configure device management policies that align with the new hybrid intelligence features. This includes defining which tasks are allowed to run locally versus those that require cloud connectivity, ensuring compliance with internal security standards.
What to do and how to stay safe: Microsoft
- Review Microsoft’s documentation on Copilot permission requests to understand how the assistant seeks approval for file and setting changes.
- Audit local file structures to ensure sensitive data is not inadvertently included in the context provided to on-device AI models.
- Configure device management policies to restrict or allow specific local AI tasks based on organizational security requirements.
- Monitor for the release of official updates for Copilot+ PCs and test the new features in a controlled environment before wide deployment.
Step-by-step guide: Prompt Injection Warning Signs: How to Spot AI Manipulation
General security guidance from the Threat Times newsroom. It is not confirmed advice from the organisations named in this story.
Frequently asked questions
When will these new Copilot features be available?
Microsoft stated that the three new capabilities will be available to Copilot+ PC users in the coming months.
Can Copilot change system settings without permission?
No, Microsoft executive Jacob Andreou stated that Copilot has the same control level as a user but always operates with permission.
Will Copilot always use the cloud for tasks?
No, Microsoft announced that certain tasks will run on local AI models, with cloud access reserved for harder tasks or when specifically needed.
