Microsoft Teams Now Lets You Turn AI Off Mid-Meeting

Microsoft shipped a toggle this month that lets Teams meeting organizers turn AI off in the middle of a call.

Copilot, Facilitator, intelligent recap. All of it, off with one switch, while the meeting is still running. The rollout started in early July and wrapped by mid-month across Windows, Mac, mobile, and web.

Most of the coverage framed this as Microsoft caving to a user revolt. That's accurate. It's also the least interesting part.

For about two years, the only question anyone asked about AI meeting features was whether they worked. Could it summarize the call. Could it catch what you missed. Nobody spent much time on when it should be running.

Microsoft answered that one for you. The answer was "always."

What a mid-meeting switch admits

Think about what this feature implies. Somewhere inside a normal call, the conversation changes character. You open with a project update, and forty minutes later someone mentions a health issue, a client's numbers, or a person who might get let go.

The recap doesn't know the difference. The transcript doesn't either.

A toggle at the top of the meeting would have been a settings change. A toggle you can hit at minute forty is Microsoft conceding that a meeting is not one thing, and that the decision about what gets captured cannot be made in advance by an admin who isn't in the room.

Here's where I want to slow down.

The toggle puts the decision in exactly the right place and exactly the wrong moment. Only the organizer knows when the conversation turns. But asking someone to notice the turn, remember the switch exists, find it, and flip it while a colleague is mid-sentence about something sensitive is asking a lot of a person who is also trying to run the meeting.

Most people won't manage it. They'll remember afterward, which is the wrong time to remember.

Decide in advance, not in the moment

So don't lean on the toggle for the things you can already predict.

Pull up your recurring meeting list this afternoon. Not every meeting, just the standing ones. Mark the calls that should never have AI running: one-on-ones, performance conversations, anything with a lawyer on it, anything where client confidential numbers get said out loud.

Then handle those with your admin at the policy level instead of at the memory level. Microsoft's own note points out the toggle doesn't even appear when Meeting AI is disabled by policy. For meetings you already know are sensitive, that's the cleaner answer.

Policy is for the patterns. The toggle is for the surprises.

The part that isn't about privacy settings

There's a second effect here worth naming, and it has nothing to do with compliance.

When AI runs by default in every meeting, people talk differently. Not dramatically. Just slightly. The half-formed idea stays unformed. The "I'm not sure about this yet" goes unsaid. Nobody announces that they're self-censoring in front of a recap bot. They just do a little less of the thinking out loud that meetings are supposed to be for.

I'm not going to tell you that's a crisis. There's no good data on it, and I'd be making it up if I put a number on it. But it's the kind of second-order effect that never shows up in a productivity dashboard, which is exactly why it survives so long unnoticed.

The cheapest fix costs you one sentence. At the start of your next call, say whether AI is on. Not a legal disclaimer, just "recap's running" or "I've turned it off for this one." People calibrate immediately. You'll also learn fast which meetings your team actually wants it in, because they'll start telling you.

That's worth more than any setting. A tool that captures everything gives you a perfect record of a conversation people were slightly afraid to have.

Microsoft spent two years telling customers what AI would do for their meetings. This toggle is the first time the company asked what they wanted in the room.

It's a small switch. The question underneath it is not: which conversations do you actually want a machine to remember?