What to check before sharing data with an AI assistant

Typing into an AI assistant feels like a private conversation. It is closer to sending a message to a service, and what that service keeps, who can read it, and what it is used for are set by its policies and settings, not by the friendly tone of the chat. A few minutes of checking before you share anything sensitive is worth it.

What happens to what you type

Depending on the service, your messages may be:

• Stored for a period, so your history appears next time and so the provider can investigate abuse.
• Reviewed by people, for safety or quality, in some cases.
• Used to improve the product, which can include teaching future models, unless you opt out or your plan excludes it.
• Shared with other companies that run parts of the service, such as hosting or search.
• Kept in memory features, which store facts about you for later conversations.

None of these is automatically bad. The problem is not knowing which apply to you.

Questions to answer before you share

• Retention. How long are conversations kept, and can I delete them? Does deleting remove them everywhere or only from my view?
• Use for improving models. Is my content used to teach future models? Can I turn that off, and is it off by default?
• Human access. Can staff read my conversations, and under what circumstances?
• Other companies. Which outside services receive my content?
• Location. Where is data stored and processed, if that matters for your work?
• Memory. Does the assistant remember things between chats? Can I see, edit, and clear what it remembers? For how that works, see AI memory vs. context.
• Sharing. If I make a link to a conversation, who can open it, and can I revoke it?
• Account type. Personal accounts, business plans, and programming interfaces often have different terms. Check the one you actually use.

The answers are normally in the privacy policy, the terms, and the settings page. If you cannot find them, treat the answer as unknown.

What not to paste

• Passwords, access keys, tokens, and private keys. Once pasted, they should be considered exposed. If you pasted one by mistake, replace it.
• Other people's personal details. Names with addresses, health information, ID numbers, and private messages belong to people who did not agree to this.
• Regulated or contractual material. Client data, medical records, and anything under a confidentiality agreement may be off limits regardless of the tool's settings.
• Unreleased work that would hurt you if it leaked.

Ways to share less

• Redact. Replace names, numbers, and identifiers with placeholders like "Client A" before pasting.
• Share the excerpt, not the file. Paste the paragraph you need help with.
• Describe instead of pasting. Many questions can be answered from a description of the situation.
• Use separate accounts for separate purposes, and keep sensitive work out of accounts with long retention or sharing turned on.

Extra care for agents

An agent that can open files, run commands, or browse on your behalf sees more than what you type. Before connecting one, find out what it can read and change, whether risky steps wait for your approval, and what is kept afterward. The principles in why an AI agent should ask before it acts are a good yardstick, and you can run an agent in a throwaway folder or virtual machine until you trust it.

Quick answers

Is a temporary or incognito chat private? Features with names like these differ between services. Read what the specific one promises about history, improvement of models, and retention, rather than assuming.

Does deleting a chat remove it everywhere? Not necessarily. Copies can remain for a retention period or in backups. The policy should say.

Is a business plan more private than a personal one? Often the terms differ, for example about whether content is used to improve models, but it is not automatic. Check the terms of the plan you actually use.

What if I pasted a secret by mistake? Treat it as exposed. Replace the password or key, and delete the conversation if you can.

Where NAMELESS stands

NAMELESS is an AI model that is still in training. Our privacy policy is still a draft and will be published before NAMELESS leaves private beta. What is stored, for how long, and how you control it belong in that policy, and we will not describe them in detail before it is final. Until then, apply the checklist above to us as you would to anyone.

You can join the waitlist to follow progress.