Microsoft detail ‘agentic AI’ plan for Windows 11, immediately admit it might install malware on your PC

Microsoft have delivered a timely reminder that AI isn’t just effective at injecting ugly, soulless anti-art and nonsense robochat into your games – it also has the power to ruin your entire PC.

MS recently updated a support article explaining the ‘agentic AI’ features they plan to roll out for Windows 11, which involve creating a separate desktop instance and allowing AI ‘agents’ to perform supposedly menial tasks (like file sorting and email sending) within it. All while you kick back on your main workspace, online shopping for the third yacht that the AI revolution has surely enabled you to afford. Unfortunately, among the agents’ capabilities is the risk of installing malware.

Specifically, the article warns that agentic AI may “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.” Sounds awful. Terrible! Clearly not the kind of vulnerability that Microsoft could possibly allow to stand and oh no wait they’re going through with it anyway, with the first experimental versions already being live via the Windows Insider programme.

There are token safeguards: the ‘agent workspace’ is disabled by default, and can only be enabled by someone with admin privileges. But it’s hardly reassuring when the only way to use these features safely is to not use them, at all. And you would stay at effectively zero risk of XPIA attacks if you didn’t, as these types of malware are engineered to target large language models (LLMs) rather than humans.

It’s likely inevitable that Windows 11 will swell up with agentic AI at some point; Microsoft have been talking about it for months, and with Windows 10 having lost (of all things) its security updates, it’s not like fleeing to the older OS is a long-term alternative. But putting it in users’ hands when the AI involved hasn’t been sufficiently trained to avoid obvious, often plain-English phishing or malware injection scams is spectacular arrogance. Unless it’s some kind of Valve inside job to make us all buy SteamOS-powered Steam Machines, but I’m not sure their wheelie desks would roll that far.

Motoring critics often use the phrase “Car Brain” to illustrate how otherwise rational, sensible drivers suddenly lose their empathy and perception functions as soon as they sit behind a steering wheel. I’d like to propose the usage of “AI Brain” to describe the phenomena of ostensibly intelligent technologists instantly abandoning any sense of responsibility, creativity, good product design, and in many cases, basic human decency the minute they have artificial intelligence in front of their face. AI Brain can be easily identified through symptoms such as excessive LinkedIn posting and the vocalisation of gibberish about how their company needs to steal other people’s work, actually.