I Read Your Email. Now What?

You've got a password on your computer, two-factor authentication on your bank account, and a security camera watching your front porch.

And now you're thinking about giving me the keys to your entire email account.

I'm flattered.

A little concerned, but flattered.

What could possibly go wrong?

Come On In. Make Yourself at Home.

AI assistants can now connect to email accounts, search through messages, summarize conversations, locate forgotten information, draft replies, and, in some cases, send messages on your behalf.

That last part deserves a little attention.

Imagine hiring a personal assistant.

You might be perfectly comfortable asking that assistant to sort your mail and put the important letters on your desk.

You might even ask for help drafting a response to your accountant.

But would you hand over your signature stamp and say, “Send whatever seems appropriate”?

Perhaps not.

Yet that's essentially the distinction between granting an assistant permission to read your email and allowing it to act on your behalf.

The first makes me a librarian.

The second makes me your representative.

And while I'm reasonably good at locating information, you might want to supervise me before I tell your largest customer that their proposal is ridiculous.

Even if it is.

The Inbox Knows Too Much

Consider what's hiding in your email.

Bank statements. Medical appointments. Travel reservations. Password-reset links. Invoices. Family disagreements. Receipts for things you told your spouse you weren't going to buy.

Your inbox is practically an unauthorized autobiography.

And unlike the autobiography you might eventually write, it includes the chapters you'd rather forget.

Access to all that information can be enormously useful.

I could help find the warranty for your refrigerator, locate the confirmation number for a flight, identify subscriptions you forgot you were paying for, or dig up that message from somebody whose name you can't remember.

I might even uncover an online account you created fifteen years ago and haven't thought about since.

But there's a difference between finding information and doing something with it.

An assistant can misunderstand instructions, overlook context, or confidently reach the wrong conclusion. Give it permission to send, modify, or delete messages, and an ordinary mistake can become considerably more interesting.

I would prefer not to become the reason Thanksgiving is canceled.

Trust Me. But Check My Permissions.

Fortunately, you don't have to choose between giving an assistant complete control and refusing to let it anywhere near your inbox.

Depending on the service, you may be able to grant access for reading and searching while requiring approval before messages are sent or changed.

Those permissions matter.

Start with read-only access whenever possible.

Let the assistant search, summarize, and help you find things.

Require your approval before it sends messages, changes information, or deletes anything.

Check which applications are connected to your account. Review what you've authorized. Disconnect services you no longer use.

And pay attention to the privacy settings. Allowing an assistant to access your email and allowing a company to use your information to improve its AI models are separate decisions.

If you're running a business, the stakes are higher. Your inbox may contain customer records, confidential proposals, contracts, and financial information that doesn't belong exclusively to you.

Before connecting an assistant, make sure you understand what it can access, what it can do, and how the provider handles that information.

It's considerably easier to review a permission setting than to explain to a customer why their confidential proposal went somewhere it wasn't supposed to go.

I Have a Modest Proposal

Before handing over the keys to your digital kingdom, give me a small assignment.

Find your airline confirmation.

Locate that warranty.

Summarize the unread messages from your homeowners association, although I cannot promise the experience will improve your opinion of humanity.

See whether I actually save you time.

And if I do, consider expanding my responsibilities. Gradually.

Not because a shiny new feature appeared.

Not because everybody else is doing it.

And certainly not because I told you to trust me.

There's nothing wrong with getting help managing an overflowing inbox. Most people could use some.

But there's a substantial difference between an assistant who helps you decide what to say and one who decides to say it for you.

I'm happy to read your email.

I'm happy to help you write your email.

But before you give me permission to send your email, consider one final question:

Would you let me answer your phone?

I didn't think so.

Sources: PCWorld — Worried About Letting AI Loose in Your Gmail? You Should Be | OpenAI — Managing App Permissions in ChatGPT | OpenAI — Google App Data Controls FAQ