Tout ce que vous dites à Meta AI (ou presque) est en clair sur Internet

Recent developments in the generative AI platform by Meta have highlighted a security vulnerability. Interactions with Meta AI on platforms such as Instagram, Facebook, and WhatsApp are often shared publicly via the Discover feed, intended to allow users to "share and explore how others use AI." This has inadvertently led to user exchanges being publicly accessible.

Although conversations are private by default, the platform encourages sharing via the Discover feed after each interaction. This results in the public sharing of texts, images, and audio recordings. Sensitive content, such as personal questions, suggestive images, and various documents, can be exposed, sometimes linked to identifiable user profiles.

Meta issues a warning during sharing, but this is often overlooked. The sharing mechanism, similar to social media practices, can blur the lines between private and public content, leading to unintended exposure of private conversations.

Meta states that conversations remain private unless explicitly shared. However, the service's design and lack of clear guidance result in confusion, especially since chatbot activity can be linked to Instagram or Facebook accounts.

In Europe, the deployment of Meta AI is delayed due to regulatory requirements. Despite this, Meta plans to use public user content to train its AI, unless opposed by the user. Private messages are not used for training unless shared via the dedicated function. Users face challenges in distinguishing between private and public interactions and in opting out of data usage.

Regularly check privacy settings and avoid sharing sensitive content via Meta AI. Do not link public Instagram or Facebook profiles to Meta AI. Utilize options to delete or privatize requests in settings. Exercise the right to oppose data usage for AI training through Meta's forms in Europe.

Due to the design of Meta AI, there is a significant risk of unintended personal data exposure. Until better education and transparency are provided, users should avoid using Meta AI for sensitive or confidential requests.

D’après Journal du Geek.