tech
Microsoft tries to head off the “novel security risks” of Windows 11 AI agents
Agents with read/write access to your files create big security, privacy issues.
TL;DR
- Windows 11 is incorporating experimental "agentic" AI features for background task automation.
- Copilot Actions aim to provide users with an active digital collaborator for complex tasks.
- AI agents will operate with separate user accounts and monitored access to files.
- Users must approve data requests, and agent actions will be observable.
- Potential security risks include data exfiltration and malware installation via prompt injection.
- These features are currently experimental, can be toggled off, and are off by default.
- Microsoft is also making Copilot more approachable with an animated character named 'Mico' and improved voice input.