The recap nobody asked for
A team lead in Rotterdam opened her Monday call and found a summary already waiting from the last one. Nobody had turned it on. Facilitator had been listening, transcribing and recapping every meeting by default, and the notes had been sitting in a channel a contractor could read.
That quiet default is what Microsoft just walked back. After months of complaints from IT and privacy teams, it is handing organizers a switch to turn meeting AI off, and the way you use that switch is now a decision, not a setting.
What Microsoft actually shipped
The change is narrow and specific. Licensed organizers and presenters can now turn Copilot, Facilitator and Intelligent Recap on or off during a live Teams meeting, one at a time or all at once. Targeted release began in early July 2026, with general availability rolling through the month.
Before this, the AI leaned on by default and switching it off was buried. The reversal followed a loud backlash, most of it aimed at Facilitator, the feature that quietly monitored and summarized what people said.
A mid-call switch is not consent
Here is the part the headlines miss. A toggle a presenter flips halfway through a call does not create a lawful basis to process what your colleagues and clients say. Under the GDPR, recording, transcribing and summarizing identifiable people needs a basis you can point to before the meeting starts, not a checkbox someone remembered mid-sentence.
Consent, if you lean on it, has to be informed and freely given. A prompt that appears after the AI has already been listening is neither. And for anything sensitive - a grievance call, a health discussion, a deal under NDA - the bar is higher still.
The switch helps you comply. It does not comply for you. Someone in your organization still has to decide when meeting AI is allowed, who may turn it on, and how consent is recorded and kept.
The decision this forces on you
Make it a policy before general availability reaches your tenant. Decide which meetings may use AI at all, name who is allowed to enable it, and require that participants are told and agree before transcription starts. Default the tenant to off and let trusted organizers opt a meeting in, rather than the reverse.
Where employees are being recorded and summarized, bring in whoever represents them. Across much of Europe that monitoring is not the employer's call alone, and a rollout without that agreement is one you may have to unwind.
The pattern behind the reversal
This is the shape of almost every AI feature arriving in software you already pay for. It ships on, it is framed as help, and it is switched to opt-in only after enough customers push back. The base rate is not neutral: assume the next update turns on something you did not choose.
So the durable habit is not fighting each toggle. It is a standing review of what your core vendors enabled since the last release, judged against one test - would we have turned this on ourselves. If the answer is no, the default was the vendor's decision, and reversing it is yours.
Read next: Shadow AI Is Your Next Data Breach | Washington Now Sets When Frontier AI Reaches You
