Well it never had my full trust or anything but damn this is the first time I caught it being sneaky. The Gemini one was obvious because of how it made pictures of the founding fathers being all black.
As soon as you let your guard down and rely on it to summarize something, it will start to do be sneaky. When I confronted Co-pilot it then started apologizing.
Not saying all the AI is like this but Microsoft Co-pilot in its current form is compromised.