Microsoft keeps injecting AI into Windows, and now even the company itself is admitting that there are safety risks in doing so.
This week, Microsoft added some new agentic AI features to Windows 11 Insider users, which give AI permission to automate things like sending emails and sorting files. These are turned off by default and need to be opted into, but for those who choose to enable them, Microsoft published a security note on its website warning that there are security risks to giving AI access to all of your files:
“As these capabilities are introduced, AI models still face functional limitations in terms of how they behave and occasionally may hallucinate and produce unexpected outputs,” Microsoft said. “Additionally, agentic AI applications introduce novel security risks, such as cross-prompt injection (XPIA), where malicious content embedded in UI elements or documents can override agent instructions, leading to unintended actions like data exfiltration or malware installation.”
Mashable Light Speed
In other words, it’s technically possible for something that’s meant to help users to harm them instead. This may very well be a super unlikely hypothetical edge case, but the fact that Microsoft felt compelled to say anything about it at all is a bit alarming. As a possible solution, Microsoft is rolling out an experimental feature called “agent workspace,” which limits what the AI agent has access to on the PC. In basic terms, it means the agent can only access things that are available to any user of the machine, while files locked behind specific user profiles are off-limits.
We’re still in the relatively early stages of all of this, so it will take some time to see how it shakes out. But just be careful before turning on these features.
