If you run client calls on Microsoft Teams, here’s a sentence worth sitting with: until very recently, the AI notetaker was on by default, and there was no clean way to switch it off mid-call.
That means some number of confidential conversations — therapy sessions, legal consultations, HR grievances, advisory calls about someone’s money — were being transcribed and summarised by an AI that nobody in the room had explicitly agreed to.
After a loud enough backlash, Microsoft reversed course. There is now a Meeting AI toggle that turns Copilot, Facilitator and Recap off during a live meeting, and it’s reaching general availability through the second half of July.
Two clicks. But there’s a catch inside it that almost nobody is writing about, and if you have a confidentiality duty, the catch is the part that matters.
Where the switch is
The control lives in the meeting toolbar, right in the call — not buried in a settings page you’d have to leave the meeting to find.
One important limit on who can do this: only licensed organizers and presenters can flip it. If you’re an attendee in someone else’s meeting, you can see that AI is running — there’s a status indicator — but you cannot turn it off. You can only ask the organizer to.
For a therapist or a solicitor, that means one thing very clearly: be the organizer of your own client calls. If the client sends you the invite, the switch isn’t yours.
The catch: transcription turns the AI back on
This is the part buried in Microsoft’s own rollout notice, under a heading called “Transcription Dependency,” and it’s the single most important thing in this article.
Meeting AI and transcription are coupled. In Microsoft’s words: turning on Meeting AI automatically turns on transcription and generates a recap — and starting transcription automatically enables Meeting AI and recap.
Read that second half again. You switched the AI off at the start of the call. Twenty minutes in, a well-meaning colleague — or the client themselves — hits “start transcription” because they want a record. The AI is now back on, and it is summarising your confidential session.
To actually keep AI out of a meeting, both have to stay off. There is no state where transcription runs and the AI doesn’t.
That last one deserves a line of its own. Turning off Teams’ own Meeting AI does nothing to a third-party notetaker bot that your organisation has permitted. Those are governed separately, by app permissions your admin controls. If a bot with a name and a little avatar joins your calls, this toggle will not remove it.
What to tell your client
Here’s where most coverage of this stops — at the toggle. But if you have a professional duty of confidentiality, the switch is the easy half. The hard half is the conversation.
The research on this is genuinely useful, and it says something uncomfortable. In a study of patients whose clinicians used ambient AI documentation, 81.6% consented when given a simple explanation — “audio is used to help generate your note.” When they were also told about the AI features, where the data is stored, and that a company is involved, consent dropped to 55.3%.
Being fully honest costs you about a quarter of your yes’s. That’s the real price, and you should pay it anyway — because the alternative is consent that isn’t informed, which isn’t consent.
Two more findings worth carrying into the room:
- 96–98% of people said that where the audio goes, who can access it, and how it’s used were important to their decision. These aren’t details you can skip.
- Between 35% and 52% said they would hold things back — specifically on mental health, sexual health, and anything they were ashamed of — if they knew an AI was recording. For a therapist, that isn’t a privacy footnote. That’s your entire session, damaged.
So the script matters. Keep it short, lead with the choice, and make declining genuinely free:
“Before we start — Teams has an AI assistant that can take notes and write a summary. I’ve turned it off for our sessions by default. If you’d ever find a written summary useful, we can switch it on, and I’ll tell you exactly where it’s stored and who can see it. Either way is completely fine. Would you like it on, or shall I leave it off?”
That’s it. It takes eleven seconds, it names the technology, it offers a real opt-out, and it hands the decision to the person whose confidence is on the line. The research is clear that a brief, plain explanation plus a genuine choice is what people actually want — not a technical briefing, and not silence.
And if you do switch it on: tell them where the recording lives. Which brings us to the thing most people get wrong.
Where the transcript actually goes
If a Teams meeting is recorded or transcribed, the file doesn’t live in some abstract cloud. It lands in the organizer’s OneDrive (for a normal meeting) or the channel’s SharePoint site (for a channel meeting), and it surfaces in the meeting chat and the calendar Recap tab.
By default, everyone who was in the meeting can open the recording and the transcript. The organizer can narrow that to organizers only, or to specific people — but they have to actually do it.
Then there’s the layer above you: these are ordinary files in your organisation’s Microsoft tenant. They’re subject to your company’s retention policy, they’re discoverable in litigation, and administrators can reach them. For a client session, “I deleted it from the chat” is not the same as “it’s gone.”
If you’re bound by HIPAA, this is not a grey area. US health regulators treat technology that records or transcribes a session as handling protected health information — which pulls in security risk analysis, safeguards, and typically a business associate agreement with the vendor doing the transcribing. Switching the AI off is the simple path. Switching it on is a compliance project.
What this means for you
If you’re a therapist or counsellor: default to off, organise your own calls, and use the script. The self-censorship finding is the one that should decide it — a client who is editing themselves because a machine is listening is a client you cannot help properly.
If you’re a lawyer: the question isn’t just privacy, it’s whether routing a privileged conversation through a third party’s AI service creates a problem you’d rather not litigate. Off by default, and if a client asks for a summary, write it yourself.
If you’re a financial adviser: you likely have recordkeeping obligations pulling the other way. That’s fine — this isn’t an argument for keeping no records. It’s an argument for knowing whether the record is one you made or one an AI improvised, and for telling the client which.
If you’re in HR: grievance and investigation meetings are the sharpest case. An AI recap of a disciplinary conversation is a document that will be read back to you later, by someone who is not on your side. Turn it off, take your own notes.
If you’re a small-business owner: two things quietly changed. Copilot is now bundled permanently into Microsoft 365 Business Standard and Premium — the separate add-on is gone, and the price moved with it — so you are paying for this whether you use it or not. And your staff can now switch AI off in a client call. Both facts are worth ten minutes at your next team meeting.
What this can’t fix
- It isn’t retroactive. Anything the AI generated before you flipped the switch still exists, in the chat and in the recap, governed by your retention policy.
- It doesn’t reach past meetings. Every recap already sitting in OneDrive stays sitting in OneDrive.
- It doesn’t stop third-party bots. Different system, different permissions, admin-controlled.
- It can be taken away from you. If your admin has disabled Meeting AI at the tenant level, the toggle doesn’t appear at all — which is good. If they’ve mandated it, that’s a conversation with IT, not a setting you can win.
- It doesn’t get you consent. The switch handles the machine. The client conversation is still yours to have, and it’s the part that actually protects both of you.
The bottom line
Microsoft gave you an off switch because enough people said an AI listening to their meetings by default felt like surveillance. Take the win — but take it accurately.
Switch Meeting AI off, keep transcription off with it, organise your own client calls, and say one honest sentence at the top of the session. That’s the whole practice. It’s less work than the compliance memo you’d write after getting it wrong.
If you want the fuller system for using AI in confidential professional work — what’s safe to paste, what never leaves the room, and how to keep a privileged conversation privileged — our Confidentiality-Safe AI Workflow course starts free with the first two lessons.
And if you land on the other side of this and decide you want AI notes with consent, that’s a legitimate choice — our AI Meeting Notes course covers doing it properly.
Sources
- Microsoft 365 Message Center MC1319216 — Teams: in-meeting Meeting AI toggle
- Microsoft makes major AI U-turn following user revolt — TechRadar
- Customize who can access a meeting recording or transcript — Microsoft Support
- Patient and Clinician Perspectives on Ambient AI Documentation — JAMA Network Open
- Telehealth and HIPAA — U.S. Department of Health and Human Services
