We live in an age where AI can do literally anything for you, but one of the biggest questions surrounding the whole practice is whether you should actually let it. Even disregarding compelling moral arguments, AI is still in its infancy and is plenty capable of making mistakes that a human touch wouldn’t allow. It’s how you get AI chatbots saying slurs in Fortnite, and AI causing more damage than good.
That’s why when Microsoft says that it’s making AI agents for Windows 11 that can access your PC’s files and do things on your behalf, you’d best make damn sure you know exactly what the potential consequences are. According to Microsoft itself, one of those consequences could be an AI agent installing malware on your PC without you even realizing.
First reported by Windows Central (thanks Kotaku), a rather lengthy warning was recently published by Microsoft about its experimental agentic features that it’s planning to add to Windows 11 relatively soon. In this warning, it basically says that these agents can be tampered with if someone feels like it, and could potentially extract and share your data with others, or install malware without your permission.
Dispatch Director Says AI Is Only A Creative Tool “If You Aren’t Creative”
Dispatch’s creative director says that AdHoc won’t use AI voice acting, and suggests it’s only a good replacement if “you aren’t creative”.
“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,” explains the warning. “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.”
It’s a rather scary prospect, which is exactly why these agents are being kept turned off by default for now. If you want to play around with them, you’ll have to manually activate them yourself, though if you ask me, potentially putting my PC and data at risk just to use an AI agent to do some busywork for me isn’t really worth the trouble. It’s possible that they could be turned on by default down the line, but for now, your data is safe unless you put it at risk yourself.
