Anthropic Puts Claude Inside Live Video Meetings
Anthropic has launched Claude Meetings, a feature that brings its Claude assistant into video calls to work on tasks while participants are still talking. The feature is in beta, an early test release, on Team and Enterprise plans. At launch it works with Google Meet, according to TechCrunch.
Claude joins as a visible participant. A team adds it to a call, and it identifies itself as a meeting participant and posts a notice in the meeting chat. Once inside, it follows the conversation and works on tasks in real time. Participants can ask it questions through the meeting chat, and it can retrieve information from connected work apps, the third-party tools a company has linked to its account, during the call.
The feature targets the planning part of a meeting. Claude can draft the plan under discussion as participants talk, and it can map out a timeline on a whiteboard while the discussion is under way. The output does not stay inside the call. After a meeting, notes, action items, and other materials Claude created are available in claude.ai.
Access to external participants is gated in two ways. Claude will not join calls that include external guests unless an admin, the account manager who controls settings, authorizes it. If an external guest joins a call that Claude is already attending, Claude automatically leaves.
The broader context here is the question of who is in the room. A meeting assistant with access to connected work apps is only as safe as the boundary around it, and at launch that boundary is set at the administrative level, through the admin authorization requirement and the automatic exit on external arrival. Enterprise security teams will want to know how those controls behave across guest scenarios, how connected-app permissions are scoped per meeting, and where the retained notes and drafts sit within their data-governance model. Those answers will shape how far such tools move into sensitive meetings such as legal, HR, or board-level discussions.
In my view, the more interesting shift is in what the assistant does rather than where it sits. Most meeting tools record and summarize after the fact. Claude Meetings is designed to produce artifacts while the conversation is still moving, so a timeline or plan can be corrected in the room instead of reconstructed from notes the next morning. That changes the cost of a bad assumption, because it can be challenged before people leave the call.
Taken together, the feature is also a test of whether an AI agent can be a useful, visible colleague rather than a silent bot. Declaring itself in the chat and appearing as a named participant are small design decisions, but they address the consent problem that has trailed automated note-takers. Beta feedback on the posted notice and admin gating will show whether teams find them sufficient.
For teams already working in Google Meet on Team or Enterprise plans, the practical question is narrow: does a Claude-generated plan and timeline shorten the gap between discussion and follow-up?
In my view, if it does, the value will be measured in action items closed, not in features demonstrated.
