Here's how Meta AI "leaks" your private chats, thanks in part to its terrible UX

Back in April, Meta launched its standalone Meta AI app, its answer to apps like ChatGPT. Powered by Llama, its own advanced AI models, the app was meant to be your conversational partner. It can write poems, help you brainstorm, generate images, and even remember details about you. The problem is, it also has a feature that unintentionally helps you "leak" your most intimate, embarrassing, and legally risky conversations onto a public feed for the world to see.

This entire mess centers on a "Discover" tab baked into the app. A feature that feels more at home on Instagram or TikTok has been bizarrely grafted onto a tool some people are using as a private diary. At first glance, the feed shows what you might expect: people generating funky AI images or asking for creative writing help. But scroll for more than a minute, and you will find some deeply personal stuff.

Neowin, while reviewing the app, saw people posting chats discussing their extramarital affairs and sharing private medical records. In one chat (click to expand the screenshot below), a user asks the AI, "is it bad if my vagina smells?"

In another, someone seeks advice on how to handle red bumps on their inner thigh. One user, grappling with a gift for a woman in a "troubled casual dating relationship," types out his entire romantic dilemma. And here's one, where a user confesses their love for their wife who is several miles away, but admits they're developing feelings for someone else:

There is also some genuinely heartbreaking stuff. An elderly, disabled dancer recounts being shooed off the dance floor at a senior community event. "I was approached by someone and was told that I couldn't be on the dance floor," they shared with the AI.

Some posts even contain voice notes, so you can hear the actual stress and sadness in people's voices as they, for example, try to figure out how to bring their deceased brother, a US Army veteran, home from a foreign country.

The feed is also a goldmine for anyone looking for legal or financial drama. One person asks for advice about their sister, a VP at a small company that has not paid its corporate taxes for more than a decade. They wanted to know if the sister would be liable:

For this record, this is what the comments section on that one looks like:

Here's another example from an accidentally public chat, where a user asked one of those "tell me everything you know about me"-style questions. Meta AI obliged, spilling out a detailed profile of the man: name, age, hometown, business aspirations, and personal goals, all pulled from past conversations:

The root cause of this privacy dumpster fire is Meta's baffling user interface. At the top right of every chat, there is a "Share" button. It is not clear where you are sharing to. There is a small, easy-to-miss warning at the bottom of the screen to avoid sharing personal information, but the path to public humiliation is incredibly simple. You tap "Share," are shown a preview screen where you can edit a title, and then you tap "Post." That is it. There is no friction, especially for first-time users, no big, attention-grabbing banner that says, "Hey, this is going public for anyone using this app to see, comment on, and remix."

Other platforms, like ChatGPT and Gemini, handle this responsibly. They let you share chats by generating a link you can choose to send to specific people. They do not just hurl your conversation into a public square. Also, enjoy this chat involving someone arguing with Meta AI, asking it to explain why there's so much widespread accidental sharing of personal information (click to expand):

The most infuriating part is that this is a problem of Meta's own making. When anyone signs up for the app, there is no prompt. No pop-up about privacy settings, no warning that their chats could end up on a public feed. You are just dumped straight into the app, completely blind to how sharing works.

The result is a whole community of users trying to clean up Meta's mess. If you’ve been paying attention, the comment sections are full of people warning others that their private thoughts are now out in the open. Some of the original posters, after being alerted, even jumped in to thank them and claimed they had no idea their chats were public.

If you are one of the people who use the Meta AI app, you are probably wondering how to clean up this mess. Luckily, there is a way to claw back some privacy, though you should not have to do this in the first place. You will want to do two things: hide everything you have already shared publicly, and then stop Meta from suggesting your future chats on its other apps.

Here is exactly how you do it. To make all your past public prompts private again:

1. Select your profile on the top right to go to settings.
2. Tap "Data and privacy."

   

3. Next, tap the "Manage your Information" button

   

4. Then, tap "Make all public prompts visible to only you"

   

5. Finally, use "Apply to all" on the confirmation pop-up.

   

Next, you will want to stop Meta from suggesting your prompts in its other apps. Neowin's digging uncovered this:

1. In settings, under Data and privacy > Manage your Information, select "Suggesting prompts."
2. Deselect "Facebook" and "Instagram."

Let's just hope Meta fixes this before something more serious happens, like a government official accidentally leaking war plans (heh). Neowin has reported the problem through the Meta AI app's own "report a technical issue" feedback form, complete with a link to this article. We will update you if we hear back.