Who should approve what Claude does inside your apps?

Govern Claude interactive connectors by naming people, not by trusting defaults. Before wider rollout, decide which connectors are switched on, which tools stay read-only, who approves Claude's actions inside each app and who reviews a shared sheet, whiteboard or data story. Anthropic provides the admin settings; your organisation provides the owners.

Claude no longer just talks about your work. In supported apps it now opens the page, row or file and edits it while you watch. That changes the governance question from 'what can Claude see?' to 'who signs off what Claude does?'

The timing matters too. Anthropic set Claude Data to switch on by default for Enterprise organisations on 19 November 2026, so some new capabilities reach staff unless an admin makes a different choice. If you are planning a rollout, the decisions below belong on this quarter's agenda. For more on running Claude at scale, see the Generation Digital blog.

What has changed with Claude interactive connectors?

Interactive connectors first arrived as small live views of other tools. The November update goes further. Where a partner app supports it, the app can sit beside the Claude chat or fill the screen, and Claude can edit the open page, row or file directly. You see each change land in the tool as it happens (Anthropic, 2026).

Approval works the way it already does for connectors. When Claude wants to act inside an open app, the user approves that action in the same manner as any other connector action. For Confluence and Notion, the full editor opens, and the user is signed in under their own account.

Launch partners in the directory include Asana, Canva, Carta, Dropbox, monday.com, Smartsheet, Vercel and Webflow. Anthropic also shows examples from Adobe, Atlassian, Box, HubSpot and Notion. The feature is on web and the Claude desktop app, across all plans, inside the new Claude experience that brings chat and Cowork together. It appears once an account updates or an organisation turns it on.

Under the hood, interactive connectors use MCP Apps, an open extension to the Model Context Protocol. Developer support is still in public preview. One detail matters for admins: if your organisation already allows every tool on a connector, new features show up as they roll out. You can still restrict a connector's tools in the usual way.

What controls do admins and security teams get?

The wider Claude ecosystem is large. The Claude Marketplace now offers more than 9,000 connectors and plugins (Anthropic, 2026). That is a lot of possible doors into your systems, so the controls matter.

• Who uses what. On Enterprise plans, admins choose which people can use which products and connectors, and they can set detailed spend limits.

• What each connector may do. Claude acts within the permissions a user already holds. Admins can narrow a connector further, for example letting Claude read Slack but not post to it.

• How people connect. Auto-connect, in beta on eligible Team and Enterprise plans, lets an admin configure an app such as Notion, Atlassian or Dropbox once. Staff then connect the first time they use it.

• What security can see. Security teams can review activity in audit logs. Anthropic's Compliance API links that activity to over 100 security and compliance products (Anthropic, 2026).

What about shared sheets, whiteboards and data stories?

Anthropic also launched Claude Whiteboards and Claude Sheets in beta, and moved Claude Dashboards to general availability under the new name Claude Data. All three run on paid plans, on web, desktop and mobile (Anthropic, 2026).

Default settings differ by plan. On Team plans, all three are on from the start. On Enterprise plans, Whiteboards and Sheets begin switched off, and Claude Data was set to switch on by default on 19 November 2026. Admins can change any of this in organisation settings, under Artifacts.

These artifacts are built to be shared, which is why they need a reviewer:

• Claude Data live sharing (beta) lets viewers see the output of queries someone has approved, even without access to the underlying connector.

• Certified metrics (beta) let a data team mark queries as trusted.

• Tracked changes in Claude Data allow a rollback.

• Claude Sheets outlines each edit so changes are easy to follow. For now it offers simple spreadsheet functions.

What is Generation Digital's view?

Our view: in-app approval only works if someone owns it. Before switching on interactive connectors beyond a pilot, name an approver role for each connected app, decide which connector tools stay read-only, and name who reviews a shared artifact (sheet, whiteboard or data story) before it goes wider. The settings exist; the ownership decision is a people decision.

This is a point of view, not a finding from client work. It follows from how the product is designed. Approval happens at the moment of action, by the person in the chat. That person needs to know whether they are allowed to approve, and someone needs to have decided that in advance.

What does this mean for you?

Treat the rollout as a set of owned decisions. The table below gives each step a named role. Adjust the titles to match your organisation.

Start with one team and one connected app. A short pilot with a clear approver tells you more than a broad launch with none.

What should you avoid?

• Leaving 'allow all tools' in place by accident. If a connector already permits every tool, new features arrive as Anthropic ships them. Decide whether that suits you.

• Assuming the user in the chat is the right approver. The product asks the user to approve. Your policy decides whether they should.

• Sharing data stories before anyone checks them. Live sharing reaches people without connector access, so review the queries first.

• Missing the Claude Data default. Enterprise admins should check the Artifacts settings rather than discover the change from staff.

Want help setting approvers, scopes and reviewers before your rollout? Book a Claude briefing with Generation Digital. You can also read more of our Claude guidance on the Generation Digital blog.

FAQ

What are Claude interactive connectors?

Claude interactive connectors let a supported partner app open inside Claude, beside the chat or full screen. Claude can then edit the open page, row or file directly, and the user sees each change appear in the tool. Anthropic lists launch partners such as Asana, Canva, Dropbox and Smartsheet. It works in the browser and the Claude desktop app, on every plan, within the new Claude experience.

Who approves Claude's actions inside a connected app?

The user in the conversation approves Claude's actions inside an open app, in the same way they approve any connector action. The product handles the approval step, but it does not decide who should be allowed to approve. Generation Digital's view is that each organisation should name an approver role for every connected app before rolling interactive connectors out beyond a pilot.

Can admins stop Claude editing data in a connected app?

Yes. Claude works within the permissions each user already has, and admins can narrow what a connector may do, for example allowing Claude to read Slack without sending messages. Where every tool on a connector is already permitted, Anthropic's new capabilities arrive automatically, so admins should review tool scope before wider use rather than relying on earlier settings.

Is Claude Data switched on by default for Enterprise?

Anthropic set Claude Data to switch on by default for Enterprise organisations on 19 November 2026. On Enterprise plans, Claude Whiteboards and Claude Sheets start switched off. On Team plans, all three are on by default. Admins can change these settings in organisation settings under Artifacts, so check them and decide who reviews shared artifacts before staff start sharing.

How can security teams monitor Claude connector use?

Security teams can review activity through Claude audit logs. Anthropic's Compliance API connects that activity to over 100 security and compliance products (Anthropic, 2026). Enterprise admins can also control who uses which products and connectors, and set spend limits. A named security lead should confirm the feeds reach existing monitoring tools before connectors go beyond a pilot.

Last updated: 11 Nov 2026

By Ollie Neagus, AI Solutions Consultant, Generation Digital