I have used Windows from 3.01, WFWG, 95, 2000, XP, ME, 7, 10, and now 11. I think that Microsoft hit its peak in 2000-XP. Now, I don't trust Microsoft with anything. I don't trust Microsoft Dropbox for anything more personal than cat photos. I switched to Linux Mint for my daily drivers, and use Windows with a carefully restricted data set. In the end, it's MY data, Microsoft, not yours.

Once again a reminder of the risk of using Microsoft software (and many other, but not all, proprietary systems). Now it's an "unspecified code error", so I'll speculate a bit, but there's plenty of history here, so we can guess the truth.

There are ways for external software to interface with Exchange / Office 356. In fact, AI systems could happily be built to work directly with IMAP and other standardized interfaces. That's what Microsoft would expect external software companies to use and they would almost always mean that simple correct configuration of the mail server would stop that software being able to see the contents of these mails at all.

Instead of giving their own software the standard interfaces and allowing everyone who uses it to have control in the standard ways, Microsoft wants Copilot to have an advantage over the competition. They allow non standard, special interfaces for Copilot whilst the competition have to stick with the standards and suffer slower development. The user suffers from more complexity (two separate interfaces), more bugs and, like here, total loss of control and security. Microsoft sells out the user for more of their own sales.

so they're saying the software is bugged ?

"Here's a summary of all your confidential emails. It would be a shame if this all got out somehow."

It is a feature, not a bug!

But hey, lets claim incompetence and not malice even though they are indistinguishable at this point.

You can't run badly tested software relying on alpha level technology (AI) and have any reasonable expectations it's going to work properly. Why are people doing this? There's a reason AI is being banned in any workplace with NDAs, trade secrets, and customer data.

All the CEOs and MBAs out there mandating that their employees use AI daily as much as possible, despite only being alpha testing level, are idiots.

This comes from the company who JUST suggested AI is ready to take away pretty much ALL white collar jobs within the next 18 months.

Sounds like this means that DLP isn't integrated at the lowest levels but is just a bolt-on thing that's advisory at best. Makes me wonder if this also means cross-tennant protections aren't as robust as one would assume.

The AI sucking down all your confidential data accidentally let you know it's been scarfing down all your data. We'll try to fix it so it keeps your confidential emails from returning a summary to you while it continues to scarf them down going forward. Sorry 'bout the mix-up.

How is this different from Outlook previewing the first few lines of an email?

Does outlook send emails off to a remote server to generate the summary or is it locally generated?

Can anyone other than the intended recipient of the email see the AI-generated summary of confidential emails?

IF, as I suspect but don't know, the local PC generated summaries are only viewable by the proper, logged in recipient of the confidential email, what's the problem?

Now, if confidential emails are flying out to off-site MS AI

If they could break the confidentially THIS easily and amateurishly, those emails were never actually confidential in the first place.

I am wondering in how much legal trouble that puts M$, especially over in Bxl.

The three letter agencies are very pleased.

Huge time saver!

so the label of confidential is THE protection mechanism... not actually blocking access, but relying on the external tool to READ and ADHERE to the label?

This is akin to writing a book with classified info, putting classified on the page and trusting who ever reads the book skips the pages that say confidential.