Imagine your most private internet searches appearing online for anyone to see. For some users of Meta AI, this might be happening without them fully realizing it. Prompts submitted to the AI tool, along with the AI’s responses, are being posted on a public feed, raising significant privacy concerns.
This situation has been flagged by internet safety experts as potentially “a huge user experience and security problem.” The issue is compounded because some of these public posts can be easily traced back to users’ other social media accounts, like Facebook or Instagram, via linked usernames and profile pictures.
Sensitive Queries Exposed
This means individuals could be inadvertently broadcasting details about their searches, including highly sensitive or personal queries. Examples found on the public Meta AI “Discover” feed include users asking the AI for help cheating on tests, such as uploading photos of exam questions and seeking answers. One publicly shared chat was even titled “Generative AI tackles math problems with ease,” highlighting the user’s intent.
Even more personal revelations have appeared publicly, such as one user’s conversation exploring questions about their gender identity and whether they should transition.
Disturbingly, searches for generating potentially problematic images are also present. These include requests for images of characters or anthropomorphic animals wearing minimal clothing. In one instance, a search asking Meta AI to generate an image of an animated character lying down wearing only underwear was traced back to a user’s Instagram account through their username and profile picture, illustrating the direct link between the public prompt and the user’s identity.
This specific type of content request also highlights a broader challenge for Meta. The company has actively fought against the proliferation of AI tools and ads promoting the creation of explicit or “nudify” images, even taking legal action against companies behind such apps. The appearance of user prompts requesting similar explicit content on Meta AI’s public feed adds another dimension to the complex issues Meta faces in managing AI-generated material and user behavior on its platforms.
Meta’s Stance vs. User Understanding
Meta states that chats with Meta AI are private by default. According to the company, content is only shared to the public “Discover” feed if users actively choose to post it. Before a post is shared, a message is displayed warning users: “Prompts you post are public and visible to everyone… Avoid sharing personal or sensitive information.” Meta also notes that users can later choose to withdraw public posts and have settings to keep their searches private.
However, the nature of the queries appearing publicly suggests that users may not fully grasp the implications of sharing their interactions. Cybersecurity experts argue that a mismatch between user expectations of a private chat tool and the reality of a public, social-media-like feed constitutes a major problem. Users might not anticipate their private AI conversations being published publicly and linked directly to their identity, potentially leading them to inadvertently expose sensitive details.
Accessing Meta AI and the Public Feed
Meta AI, which launched earlier this year, is integrated into Meta’s popular platforms like Facebook, Instagram, and WhatsApp. It is also available as a standalone product, and it is this standalone version that features the public “Discover” feed. While Meta positioned this feed in a press release as a place to “share and explore how others are using AI,” the current examples raise questions about user awareness regarding the public and traceable nature of these interactions.
Users can adjust their account settings to ensure their searches remain private. However, the prevalence of sensitive and traceable posts on the public feed indicates this option may not be widely understood or utilized effectively by everyone.
