That makes for a funny tongue in cheek comment, but it's not MS's AI they're after, it's end user secrets, and the exploits target multiple LLMs. (by adding commands to relevant MD files)