Much of our work depends on information scattered across different tools. The discussion happens in Slack, the changes are on GitHub, and the project plan may be in a Drive document that only a few people can open. When someone needs to compare those sources, they usually open each one, piece the context together and bring the conclusion back to the team.

Anthropic's September 24 announcement aims to make that easier. Claude Tag, an assistant that participates in Slack channels, can now use the requester's personal connectors alongside the channel's shared connectors. In the company's example, it reads a plan from the person's Drive and compares it with pull requests available through the team's GitHub connection.

The appeal is clear. A question such as “what did we plan but have yet to ship?” no longer requires someone to paste excerpts from two tools into a conversation. But thinking about using this at work brings up another question: if only one person can see the plan, who will be able to read the answer built from it?

When the answer enters the channel

The personal connector follows the requester's account permissions. That governs access to the source. Once Claude posts a response, however, people in the channel can see it. These are two moments in the same task, and the move from one to the other is where a team may be caught off guard.

Consider a plan that mentions a client, a budget still under negotiation or a decision that has not been announced. Revealing sensitive information does not require copying a whole page; one sentence in a summary might be enough. To me, that is the heart of this release: gathering information takes less work, while deciding what to share becomes more important.

Anthropic offers two posting modes: the requester can review each response before it appears, or allow automatic posting with a pause when Claude detects sensitive content. The company also says Enterprise administrators will be able to require review for everyone. Automated detection cannot replace the judgment of someone who knows the context, so I would begin with review enabled in channels that use restricted documents.

Three decisions before rolling it out

For a technology leader or the person designing these workflows, I would look at three things:

  • Who can retrieve the source? Personal connectors use the requester's access; channel connectors use the permissions set up for the team.
  • Who can read the answer? Before posting, consider how it reads to a channel member who cannot open the original document.
  • Who can act when nobody is asking? Scheduled tasks and work Claude starts on its own still use shared connectors configured by an administrator. Personal access does not become a permanent automation credential.

There is a useful audit distinction too. According to Anthropic, personal connector activity is logged under the user, while the channel's connector activity remains tied to its service account. That helps teams understand which access was used at each step.

The rollout has begun on Team plans, with Enterprise expected later. I can see this saving time on work teams do every day. Before rolling it out widely, I would try it with a restricted document and read the response as a teammate who cannot access that source. If the answer helps the team without carrying over more information than necessary, this is a meaningful improvement. If it does not, review needs to be part of the workflow from the start.

Sources: Anthropic's announcement about personal connectors and its official guide to connector permissions.