This New ChatGPT Scam Could Ruin AI Forever

You're at Risk of Losing All Your Sensitive Info by Using ChatGPT

Imagine you log into ChatGPT one day to do some busy work...
And next thing you know — your computer is hacked.

Sounds scary, but this is now a possibility.

I normally talk about all the benefits of using AI, but let's dive into the dark side for a second.

In last week's newsletter, we talked about the new operator mode — also known as agent mode.
This new feature can connect ChatGPT to your Google Calendar, Google Drive, Gmail, Dropbox, and more.

Here's One Way It Can Help

Let’s say a customer emails you. Instead of:

• Reading it manually

• Asking ChatGPT to generate a response

• Then copying and pasting it into your email to send out

You can now:

• Go into ChatGPT

• Have it read the email and generate a response on its own

• Then send it on its own

It's a minor change, but it's making everything so much more streamlined and convenient.

Now Here's Where the Bad News Comes In

There’s a new type of scam that you need to be careful of...
It’s called prompt injection.

Why Is This Dangerous?

On agent mode, ChatGPT can:

• Search the web

• Log in to websites

• Save login details

There are shady websites out there where people are inserting secret prompts to override ChatGPT.
They can then pull sensitive info that you have stored on your account.

And if you're not careful, that could end up really bad.

Best Practices So You Don’t Lose Everything

• Don’t put sensitive info or upload sensitive files

• Go to Settings → Security → Turn on Multi-Factor Authentication

• Go to Settings → Data Control → Turn off 'Train Model for Everyone' and delete all remote browser data

• When using agent mode, multiple third-party apps (aka connectors) can be selected. Only select the ones you're actually using

This last step is totally optional, but consider upgrading to the Teams plan. It's $25/month/user and you need at least two users. But it comes with enterprise-level data protection. (Let me know if you'd like me to dive deeper into this.)

With all this being said, I’m still really excited to see all the new ways we can now use ChatGPT.
I think OpenAI is headed in the right direction with this new update.

But nothing is perfect.

Make sure to follow these best practices and stay safe.