Meta Is Not Scraping Your DMs to Train Its AI

I’ve seen a number of videos on my feeds circulating a serious, but perhaps not unbelievable, claim: The idea is that, on December 16, 2025, Meta will update its privacy policy to allow itself to delete users’ direct messages on Instagram, Facebook, and WhatsApp, all in the name of training its generative AI models. While Meta is not a bastion of user privacy, this particular claim is fortunately not true. You should not expect the company respect your data, but you also don’t have to worry about future iterations of Meta AI generating text based on your Instagram DMs.

If you’ve also seen these claims on your feeds, you know that they are often quite stark and hyperbolic. Snopes highlights one such viral Instagram post that says: “Every conversation. Every photo. Every voicemail. Fed into AI.” Users scrolling through Instagram may understandably stop to read these claims and may, just as understandably, feel concerned about their data on their various Meta accounts.

The problem is that these claims misinterpret a real change that Meta is planning for its privacy policy. Once the policy is updated on December 16, Meta is changing how it collects data from user interactions with Meta AI. In the future, data generated when using Meta’s AI products may be used to personalize the content you see on Meta platforms, as well as the advertisements placed in your feeds. Again, this is simply additional data collection for the purpose of promoting user engagement and ad relevance.

Part of the confusion is how this privacy policy is worded. Consider the following statement, which is part of Meta’s explanation of the user information it uses: “Interactions with the AI ​​on Meta and related metadata. For example, information that you or others exchange with the AI ​​on Meta, such as content and messages.” This may look like Meta is deleting your DMs to train its AI, but it does indicate the content and posts you share. with Meta’s AI is what’s fair. Simply sending messages and attachments to users on Meta platforms does not end up in Meta AI training data, but if You share these messages and attachments with the AI, SO it could.

Meta confirmed this to the media, sending out the following statement: “The update mentioned in the viral rumor is not about DMs at all, but about how we will use people’s interactions with our AI features to further personalize their experience. We do not use the content of your private messages with friends and family to train our AIs, unless you or someone in the chat chooses to share these messages with our AIs. This is also not new and does not not part of this December 16 privacy policy update.

Make no mistake, of course: if you use a Meta messaging platform, the company collects your data. In current and future privacy policies, Meta indicates that messages you send and receive, including their content as well as their metadata, subject to applicable law, may be used by Meta for a number of functions. This could include customizing Meta products to fit each user; improve Meta products in general; promote safety and security on its platforms; store, move and process your data worldwide; and process information where required by law.

Now there are categories that Meta will not use your messages, unless you share those messages with its AI. This includes using your data for analytics and business services; “deliver a seamless experience” through Meta products; personalize ads on Meta Audience Network; research and innovation for “social good”; anonymize your information; and share data with law enforcement. In some of these cases, Meta will collect metadata about your messages (perhaps things like time or location sent), but not data from the message itself.

Again, this article is not meant to flatter Meta; rather, it is about dispelling unnecessary concerns. Social media posts often misrepresent these nuanced points about privacy and security, especially when those posts escalate the drama. Understand that if you use Meta products, your data is constantly collected, but, in this case, your messages are not suddenly retrieved to train the AI.

The best thing you can do to maintain your privacy and security with your meta messages is to use end-to-end encryption (E2EE) whenever possible. WhatsApp has E2EE built in and Meta has automatically started rolling it out for Messenger, but you may need to manually start an E2EE chat for existing conversations in the app. The same goes for Instagram: Meta offers E2EE, but you have to activate it yourself. In either app, tap the chat name to check if that conversation is currently E2EE or not.