AI agent privacy checklist: 5 checks before you connect
Before connecting an AI agent, check what information it can read, which actions it can take, where your information may go, how memory and retention work, and how to revoke access or request deletion. Review the permission screen and the provider’s current policy for the exact account and feature you plan to use.
“AI agent” covers many products: a chat assistant, a browser helper, or a system connected to email, files, shopping, or other apps. Their access can differ by account, feature, plan, and settings, so a product description alone cannot tell you what your account has granted.
1. What can it read?
Look at the named account or data source in the connection screen. Does the permission cover one file, a selected folder, your inbox, your calendar, or a wider account? Check whether it applies once or stays available after the current task. If the screen is vague, pause and find the provider’s permission details before proceeding.
2. What can it do?
Separate reading from acting. Sending a message, making a purchase, changing a record, and viewing information are different powers. Check whether the agent asks you to approve each action, whether that approval shows the exact recipient or change, and how to stop work already in progress.
3. Where can information go?
Check the provider’s notice for connected services, third-party processors, human review, and use to improve models. The answer may depend on whether you use a personal or work account, how you reached the assistant, and the controls you selected. Follow links to product-specific notices where available.
4. What can it remember?
Find out whether the product offers saved memory, chat history, activity history, or connected data. These can be separate controls. Check where each is managed and whether deleting one item also affects saved memories or information kept for security, legal, or backup purposes.
5. How do you leave?
Find the steps to disconnect the app, revoke its permission in the connected service, delete saved information, and delete the provider account if you choose. These are different actions. Check how the provider confirms a request and what exceptions its current notice describes.
A quick pre-connection test
- Can you name exactly which data source is being connected?
- Can you tell which actions are allowed and when you will be asked?
- Can you find the memory, retention, and deletion controls?
- Can you revoke the connection without deleting your whole account?
If any answer is unclear, wait and read the current permission and official policy first.
What Privacy Pal can and can’t tell you
Privacy Pal offers learning missions and helps people interpret evidence they choose to provide. A policy page is the provider’s statement; it does not reveal your current account settings or prove how a particular agent behaved. The public agent directory is based on a supplied list and marks listings as unreviewed.
Official starting points
- OpenAI Privacy Policy and the product’s current account controls.
- Anthropic Privacy Center for consumer and commercial product guidance.
- Google Gemini Apps Privacy Hub, including connected apps, settings, and retention topics.
- Microsoft Privacy Statement, with product-specific Copilot information.
This checklist is general information, not legal advice or an independent security audit. Provider terms and product settings can change; use the linked official pages for current details.