Microsoft just gave Teams an agent that answers questions nobody asked it to

24 September 2026

A change to Microsoft Teams rolled out this month is worth flagging because it's a neat example of exactly the kind of decision a business needs to actually make on purpose, rather than let happen by default.

What's changed

Teams' meeting agent, Facilitator, launched last year mainly for note-taking and agenda tracking, has gained a new capability. When a factual question comes up in a meeting and nobody answers it, Facilitator can now run a web search through Copilot and post the answer straight into the meeting chat, without anyone asking it to. It requires Microsoft 365 Copilot Premium, has to be switched on manually per meeting, and Microsoft says it's designed to fire less than once per meeting so it doesn't become a distraction.

The detail that actually matters

Here's the part worth sitting with. Once Facilitator's enabled, its answers appear in the meeting chat for everyone present, including external and cross-tenant attendees, people who may not hold a Copilot licence themselves and have no way to know the answer they're reading was generated automatically rather than confirmed by someone in the room. Microsoft's own guidance to admins flags this directly, and suggests deciding as a matter of policy whether it should be on by default, rather than left to whoever happens to be running a particular call.

Why this is a genuinely useful test case

We've spent the last two weeks on this blog talking about AI acceptable use policies in fairly abstract terms, approved tool lists, data classification, human oversight. This is a concrete example of exactly the kind of thing those policies need to actually name. Facilitator isn't a chatbot someone chose to open. It's a feature quietly available inside a tool your whole business already uses every day, and once it's switched on, an AI-generated answer can land in a client-facing meeting, unverified, without anyone specifically choosing to share it.

For a regulated firm, or any business having a sensitive client conversation, an incorrect AI-generated answer appearing mid-meeting, in front of an external attendee, isn't a hypothetical. It's a feature Microsoft shipped this month.

What we're doing at Maple

At Maple, we won't be enabling this feature for our clients. We're raising it because it's a useful example of something businesses should be aware of and consider as Microsoft and other technology providers continue adding AI capabilities to the tools people already use.

That doesn't mean every new AI feature is something to avoid. It means understanding what has changed, where it can be used, who can see the output and whether it fits with your organisation's existing security and AI policies before switching it on.

For us, that's part of the wider conversation we have with clients around Microsoft 365, cybersecurity and responsible use of AI. Sometimes the most useful advice is simply knowing a new feature exists before someone starts using it.

What to actually do

• Add named meeting AI features, not just chatbots, to your AI tool inventory and acceptable use policy
• Decide centrally whether Facilitator's proactive mode should be enabled at all, rather than leaving it to whoever's hosting a given call
• If it's enabled for internal meetings, consider whether it should be switched off by default for anything client-facing or externally attended
• Treat "a new AI feature shipped inside a tool we already use" as a recurring governance question, not a one-off decision made once and forgotten

Small feature, useful reminder. The AI questions aren't only about which new tools to approve, they're increasingly about which existing tools quietly grew new capabilities while nobody was looking.