Meta AI in WhatsApp: switch it off, object, protect your chats
The icon in the chat window cannot be removed, the chain letter about AI reading your chats is wrong, and you can still object to AI training. What Meta AI can see in WhatsApp and what you can do about it.
This article is an overview based on vendor information, documentation and publicly available reporting, not a scored test of our own. Prices and features change quickly; the vendor's own terms always take precedence. We only score what we have used ourselves for weeks: those tools are listed under AI tools compared.
In the messenger since March 2025
Meta switched on its assistant Meta AI in 41 European countries on 20 March 2025, in WhatsApp as well as in Instagram, Facebook and Messenger. In WhatsApp it appears as a chat of its own, as an icon above the chat list and in the search bar; in groups it can be summoned with "@Meta AI".
Meta AI answers questions, drafts messages and searches the web. Technically it is a chatbot like any other, with one difference: it sits in an app that almost everyone in Germany uses, including people who would never have deliberately tried an AI service.
Meta AI cannot be switched off
There is no switch that removes Meta AI from WhatsApp. Germany's consumer advice centre puts it plainly: you cannot deactivate the assistant, only ignore it. If you never write to it, you send it nothing.
You can delete the chat with Meta AI: swipe left in the chat list or long-press it and choose "Delete chat". That tidies up the interface but does not undo anything. Whatever you have already written to Meta AI is with Meta. The icon itself stays.
What Meta AI can read, and what it cannot
Your personal chats remain end-to-end encrypted, including those with people who use Meta AI. Meta AI only sees what is sent to it: messages in the chat with the assistant and, in groups, messages in which someone mentions "@Meta AI" or that someone forwards to it. The rest of the group conversation stays closed to Meta.
The boundary has a gap that no technology closes: any group member can quote, forward or take a screenshot of a message and show it to the assistant. WhatsApp introduced AI summaries of unread chats in the United States in June 2025; in Germany they are not generally available as of September 2026.
Advanced chat privacy: the switch for each chat
Since April 2025 WhatsApp has offered "Advanced chat privacy" for every individual and group chat. When it is on, three things are no longer possible: exporting the chat, automatically saving media to the gallery and using messages from the chat for AI features, which includes mentioning Meta AI in it.
To switch it on, tap the name of the chat or group at the top and activate "Advanced chat privacy". For family groups, parents' chats and anything work-related it is a sensible default. It does not, however, protect against screenshots.
The chain letter: what is true and what is not
In June 2026 a chain letter circulated on WhatsApp warning of a deadline: "from Saturday" the AI would gain access to all chats, phone numbers and personal data unless you switched on advanced chat privacy. An older version was already doing the rounds in summer 2025.
The fact-checkers at ZDFheute, Correctiv and Mimikama, as well as the Lower Saxony consumer advice centre, rated the message false or misleading. There is no deadline, and Meta AI does not get automatic access to chats; Meta AI has been in WhatsApp since 2025. Only the advice at the end is right: advanced chat privacy does no harm. You should still not forward the chain letter.
Objecting to AI training
Since 27 May 2025 Meta has been training its AI on public content from adult users in the EU: posts, photos and comments on Facebook and Instagram as well as conversations with Meta AI. Meta excludes private messages and minors' accounts. According to Germany's Federal Data Protection Commissioner, conversations with Meta AI in WhatsApp are not currently used for training.
You can object at any time, using one form each for Facebook, Instagram and WhatsApp; the consumer advice centre links to all three on its Meta AI page. No reason is needed. However, an objection filed after 26 May 2025 only applies to the future: what has already gone into training is not taken back.
Legally, the dispute is settled for now. The Higher Regional Court of Cologne rejected an urgent application by the North Rhine-Westphalia consumer advice centre on 23 May 2025, and the Irish data protection authority had approved the plan after improvements. According to reports, the fact that Meta has also been analysing conversations with its AI for personalised advertising since December 2025 does not apply in the EU.
ChatGPT and other assistants in WhatsApp
Meta had banned general-purpose AI assistants from other providers from WhatsApp's business interface as of 15 January 2026; ChatGPT and Microsoft Copilot had to leave the messenger. On 9 June 2026 the European Commission ordered Meta, through interim measures, to reopen access, and since mid-July ChatGPT has been back in the European Economic Area.
For data protection this changes little: whatever you write to ChatGPT via WhatsApp goes to OpenAI and is subject to its rules. If you want an AI for personal questions, you are better served by an assistant with clear privacy settings, such as Claude or Le Chat, which came out ahead on data protection in our test. More options are in our article on ChatGPT alternatives.
What we recommend
Use Meta AI only for harmless matters, never for health, finances or information about other people. Switch on advanced chat privacy in groups with sensitive content. If you do not want your public posts to go into training, object on Facebook and Instagram, sooner rather than later. Do not forward chain letters.
A stricter rule applies to companies: anyone using WhatsApp for work needs a clear legal basis for that in any case, and business data does not belong in an AI chat without a data processing agreement. The checkpoints are listed in our article on the GDPR and the AI Act.
In short: Meta AI cannot be switched off in WhatsApp, only ignored; its chat can be deleted. Private chats stay encrypted, and Meta AI only sees what is sent to it or tagged with @Meta AI. Advanced chat privacy blocks that for each chat. The chain letter is false. You can object to training at any time, but only for the future.
Tools discussed in this article
Each tool has a full review with scores and pricing.
Jean-Marc Mihoc works with AI tools every day and tests them here on real tasks, with accounts he pays for himself, rather than on demos. He writes down what comes out of it, even when that is unspectacular, and says for every tool who should not bother with it.
