Eagle Harbor AI
Free guide

What's safe to connect?

A plain-English guide to Claude and your business data · Eagle Harbor AI

Connecting your everyday tools (email, calendar, files, bookkeeping) is where Claude stops being a clever chatbot and starts being useful. It's also where many people are most thoughtful about what happens when you connect something. That's a good thing! Here's a run down, so you can decide if it's right for you.

Three things cover most of it

  1. Connecting lets Claude look things up when you ask. Your email and files stay where they are, and nothing is copied somewhere new. Only the conversation itself is kept. Delete the chat and that's gone too, off Anthropic's systems within 30 days. You choose what Claude sees, and you can turn that off any time.
  2. Check your training setting once. Two minutes, and it's the only homework on this page.
  3. When in doubt, leave it out. Start with the low-stakes tools, and add the rest as they earn it.

What actually happens when you connect

You sign in on the tool's own screen, the same "Sign in with Google" you've used a hundred times, and choose what Claude can access. From there:

  • Claude can only see what your own login can see. Nothing extra comes with it.
  • Mostly, Claude just reads. The few things it can do, like drafting an email or adding a calendar event, always start with you. In Gmail, for example, Claude can write a draft but can't hit send.
  • No connector ever asks for your password. Signing in always happens on the tool's own screen. If anything ever asks you to paste a password into a chat, that's your cue to close the tab.

"Will my data be used to train the AI?"

  • On a business plan (Claude Team or Enterprise): no. The one exception is a reply someone on your team rates with a thumbs up or down.
  • On a personal plan (Free, Pro, or Max): only if you've said so. Training happens when the "Help improve Claude" setting is on, or when you rate a reply with a thumbs up or down. It lives under Settings → Privacy. This is the two-minute homework from up top. And even with it on, it covers your chats, not what's inside your connected tools (unless you paste that into the chat yourself).

A traffic light for your data

🟢 Green · connect with confidence: your calendar, your website and marketing content, brand assets, and anything you'd happily hand a trusted contractor.

🟡 Yellow · fine, with the basics in place: email, Google Drive or Microsoft 365, QuickBooks, your customer list. The basics are the training setting from above (off, or a business plan), plus a moment's thought about whose information lives in that account. Your customers are trusting you with it.

🔴 Red · keep out of any AI tool: passwords and login codes, Social Security numbers, clients' medical information, and anything a contract or NDA says stays private.

One nuance worth knowing: with the right agreements in place, health information can be handled safely. That's a conversation, so ask us.

Changing your mind takes a minute

In Claude: Settings → Connectors → Disconnect. If you'd like the extra certainty, revoke it on the tool's side too (for Google, that's myaccount.google.com/connections).

Not sure where one of your tools lands?

Bring your list to a session. Sorting that out together is exactly what we do.

Bring your list Start wherever you are.

Claude is made by Anthropic, an AI-safety company: independently audited, your data encrypted whether it's moving or stored, and employees can't read your conversations outside narrow, documented cases (like a safety review). The details are public at trust.anthropic.com.

Prepared by Eagle Harbor AI · every claim checked against Anthropic's published policies, August 2026 · questions: hello@eagleharborai.com