Facilitator will warn about meeting scams — is that the same as Teams phishing detection, and should Security rely on it?

You’re right to separate “AI that watches the chat” from “AI that watches the meeting.” They’re related Steady Security themes — and they are not the same control.

Here’s what’s going on. Roadmap 573450 (Facilitator’s enhanced protection skill can alert you to potential scams) is In development, with general availability targeted for November 2027 (Worldwide; Desktop, Mac). The MRC description is specific: when Facilitator is in your meeting, its Protection skill uses AI on the meeting transcript to look for signs of social engineering by people outside your organisation. If it suspects an attempt to trick someone into sharing sensitive information, you may see a warning in the meeting, with options to end or to temporarily restrict information sharing. That is months further out than messaging phishing (575288, November 2026 intent for conversation-context warnings in higher-risk external first-contact chats). Don’t brief staff as if Facilitator Protection ships with tomorrow’s phishing banner.

Facilitator itself already has admin and licence reality on Learn: eligible Microsoft 365 + Teams + Microsoft 365 Copilot for the person who starts it; Allow/Block in Teams admin centre → Manage apps; Loop experiences in Teams on for AI notes; notes stored as transcript-class Loop data in the initiator’s OneDrive Meetings folder; Facilitator replies are visible to the meeting, unlike private Copilot prompts. Learn also lists current limits (for example no support in external chats/meetings for Facilitator today in the documented limitations — so treat “outside your organisation” scam detection as a roadmap capability to verify against Message Centre when it approaches GA, not as something you can flip on next week). Existing meeting hygiene still matters: lobby and anonymous-join settings, presenter restrictions, Safe Links / Defender for Teams, and user reporting of suspicious meetings or participants where Defender is licensed.

This is Third Law for hybrid trust: the easier it is for an external party to join and sound helpful, the easier social engineering rides the transcript. Steady before scale keeps process and Defender layers; Facilitator Protection is a future warning layer for meetings that already have Facilitator running — not a licence to drop the lobby.

Options that actually work:

1. Treat 573450 as a 2027 pilot item on the AI/security register. Track MRC status; don’t promise organisers a scam shield in this year’s Wave 0. Best Steady honesty.
2. Harden meeting posture now (lobby, who can present, external access policy, “never share MFA codes or bank changes in a meeting”). The behaviours Protection will later reinforce.
3. Pilot Facilitator for notes with a security brief, not as a SOC tool. Allow for a named cohort; teach that @Facilitator chat is shared; Purview playbooks for Loop notes as Learn describes.
4. Assume 575288 messaging detection covers meeting scams. Different surface (chat vs Facilitator-in-meeting transcript), different GA windows — keep both on the register as separate lines.

12 Panes action / comms: Tell Security: 573450 is Facilitator Protection for suspected external social engineering in meetings where Facilitator is present — GA November 2027 intent; it does not replace Safe Links, lobby rules or 575288 messaging warnings. Tell Teams admins: Facilitator Allow/Block and Loop are prerequisites for any future Protection skill; channel-meeting limits remain a separate product story. Tell organisers: If someone asks for credentials or payment changes in a call, stop — don’t wait for an AI banner that may not exist in your tenant yet. Leadership: Detection helps when it ships; process is the floor today.

Benefits / risks: In-meeting warnings could catch verbal social engineering that URL scanners miss; GA is far enough that dates, platforms and admin toggles will change; Protection only helps meetings that already run Facilitator; false positives and “restrict sharing” UX are undocumented while status is In development.

I’d suggest a half-page Steady note: keep your external-meeting and callback-fraud rules live, list 573450 beside 575288 on the register with owners and “re-check Message Centre before promising users,” and only plan a Facilitator Protection pilot when Learn publishes admin guidance — not when a roadmap card looks exciting. You’re asking the right question; the answer is patience with process, not waiting for an agent to save the meeting.

DIGITAL SKILL WORKS LTD / 12 PANES · You ask, we answer