Is Meta AI private? What happens to your chats in WhatsApp and Instagram

It’s right there: the blue circle in your WhatsApp chats, in the Instagram search bar, in Messenger. So you ask it things. A recipe. What a rash might be. How to word a tricky message to your boss. And WhatsApp is the app famous for end-to-end encryption, so that’s private too, right? Not quite.

WhatsApp: encrypted, until you talk to the AI

• Your chats with people are safe. One-on-one and group chats with friends are end-to-end encrypted: not even WhatsApp or Meta can read them (WhatsApp).
• Your chats with Meta AI aren’t like that. If you chat with Meta AI, directly or in a group, “Meta receives what you or others send it” (WhatsApp). As an AI and privacy adviser told the BBC: when you talk to Meta AI, “one of the ends is Meta, not your friend” (BBC).
• WhatsApp’s own advice: only send Meta AI what you’re comfortable with it remembering (WhatsApp).

What Meta does with your AI chats

• It trains on them. In Meta’s own words: “Meta uses your interactions with AIs to improve AI at Meta” (Instagram). Your private messages with friends aren’t used, unless you or someone in the chat shares them with the AI (Meta).
• People may read them. Meta can review your conversations with its AI, automatically or by hand (Meta AI Terms).
• They shape your ads. Since 16 December 2025, Meta uses your chats with its AI to personalize the posts and ads you see. Chat about hiking, get ads for hiking boots. Meta says it doesn’t use conversations about religion, sexual orientation, politics, health, ethnic origin, philosophical beliefs or trade union membership for ads (Meta). At launch, this didn’t apply in the EU, the UK and South Korea, TechCrunch reported (TechCrunch).
• It may keep what you say. Meta’s terms say its AIs “will sometimes retain and use” what you share, and tell you not to share sensitive topics (Meta AI Terms).

Can you turn Meta AI off?

Not completely. WhatsApp calls Meta AI optional, but you can’t remove it from the app (BBC). It’s only “on” when you, or someone in a group, sends it a message (WhatsApp). Two things you can do:

• Just don’t start a chat with it. Meta AI knows nothing about you until you message it.
• Turn on Restricted chat in a chat or group. Then nobody in that chat can use Meta AI features there. In a group, an admin turns it on; in a one-on-one chat, either of you can (WhatsApp).

Deleting: you need a command

This is the step almost everyone misses. Deleting a Meta AI chat the normal way only removes it from your phone. To delete Meta AI’s copy of your messages, WhatsApp says you have to type a command in the chat with Meta AI (WhatsApp):

1. /reset-ai resets that one chat.
2. /reset-all-ais resets all your chats with Meta AI, one-on-one and in groups.

Once it’s gone, you can’t get it back. And Meta’s terms add a warning: depending on how you use its AIs, deleting messages, threads or even your whole account “may not delete our copy” of your information (Meta AI Terms).

Asking something personal? Melo, Kamelo’s free private chat, answers on your own computer, so your question isn’t sent anywhere. How it works

In Europe, you can say no to training

Since late May 2025, Meta trains its AI on public posts and comments of adults in the EU, and on their chats with Meta AI. Meta told the Irish regulator it would start on 27 May (Meta, Irish DPC). In Europe and the UK, Meta does this on the basis of “legitimate interests”, and you have the right to object (Meta):

• Your objection covers your chats with Meta AI and what you’ve posted publicly.
• It applies to every account in the same Accounts Center. For WhatsApp, there’s a separate request.
• Object while a training run is already going, and your data may still be used for that run, but not for later ones.

You’ll find the objection forms on Meta’s page about generative AI and your information.

Instagram and the Meta AI app

• No more encrypted DMs. Since 8 May 2026, Instagram no longer supports end-to-end encrypted chats. For encryption, Meta points you to WhatsApp (Instagram).
• Mind what you type. Instagram itself says to be careful about putting your name, address, email or phone number in a message to an AI (Instagram).
• Think before you share. In June 2025, people posted chats to the public Discover feed of the Meta AI app, seemingly without realizing. Meta then added a warning before posting (Engadget).

The exception: Incognito Chat

In May 2026, Meta announced Incognito Chat with Meta AI, for WhatsApp and the Meta AI app. Meta says these chats run in a secure environment where nobody, not even Meta, can read them, and that they aren’t saved and disappear by default. It’s rolling out “over the coming months”, so you may not have it yet (Meta). Meta has published how the underlying technology, Private Processing, works (Engineering at Meta). For now, it’s the only kind of Meta AI chat built so that Meta can’t read along.

Where Kamelo helps

Kamelo doesn’t work inside the WhatsApp or Instagram apps. What it can do is give your personal questions a better place: in Chrome, Melo answers on your own computer, so your question isn’t sent to Meta or any other AI company. And when you type something personal on ChatGPT, Claude or Gemini, Kamelo suggests a temporary chat or Melo before you send. Kamelo doesn’t change your message and can’t hide who you are to your AI account.