Back to Articles
Accidental Public Sharing of Meta AI Chatbot Conversations

Crikey

READ

Details

Date Published
4 May 2025
Priority Score
3
Australian
Yes
Created
6 May 2025, 02:45 pm

Authors (1)

Description

Information about medical conditions, intimate confessions, sensitive requests and horny image generation requests are all visible on Meta's new Discover feed.

Summary

The article sheds light on the unintentional public sharing of private conversations with Meta's AI chatbot through the new Discover Feed. Users are unknowingly publishing sensitive information, including medical conditions and personal confessions, in a publicly accessible manner. This illustrates a governance failure by Meta in protecting user privacy and highlights potential risks of integrating AI capabilities into social platforms without adequate safeguards. While the content does not directly address catastrophic AI risks, it is significant in demonstrating privacy concerns and governance challenges associated with integrating AI into consumer applications. This has implications for both Australian and global AI policy in ensuring secure and ethical AI deployment.

Body

People are sharing sensitive and intimate conversations with Meta’s artificial intelligence-powered chatbot, seemingly unaware that the tech giant is making these interactions public.Questions about treating medical conditions, confessions about same-sex attraction, requests to write eulogies and generate images of movie star Sydney Sweeney in a “deep neck” dress have all been posted to a publicly available feed of AI conversations.Last week, Meta — which owns Facebook, Instagram, WhatsApp and several other social media apps —releaseda standalone app version of its artificial intelligence bot, Meta AI.Related Article Block PlaceholderArticle ID: 1195137Disabled man’s request for ChatGPT subscription denied by NDIA for ‘risk’ of false info, poor value for moneyCam Wilson9As part of the update, the social media company created a social feature: the Discover Feed, to “share and explore how others are using” the ChatGPT-style bot, which allows people to ask questions, make requests to generate content and have audio back-and-forths.While much of the feed is filled with people’s requests for AI slop-style images of “The SimpsonsinFallout 4” and people turning theirpets into cartoons,Business Insidersenior correspondent Katie Notopoulos spotted a stream of posts that people may not realise is public.Meta gives users the ability to publish their conversations with its AI bot.Some people seem to be sharing sensitive medical information as part of their questions about a condition, such as people asking how to treat peptic ulcers or for information about them.Others are publishing information like their whereabouts and movements when asking Meta’s AI to plan their trip from Townsville to Cairns.Then there’s information that, while not necessarily identifying, may be sensitive or embarrassing. One woman, whose profile links to her social media accounts, asks for help about being “really confused about my feelings for this girl that I liked for a long time”.Related Article Block PlaceholderArticle ID: 1203399Meta’s theft matters, but there are more immediate threats to Australian authorsCatriona Menzies-Pike5Another user, whose profile lists them as being from Fremantle, asks if there are any consequences for a contractor forging signatures and filing tax returns for a not-for-profit for 20 years (Meta’s AI warns of “serious consequences”).Meta’s Discover feed also includes recordings of people’s audio conversations with the AI chatbot, including long back-and-forths about people’s religious beliefs or even getting to know the chatbot.The feed also provides some fascinating insight into what people are using Meta’s AI chatbot for, including “create an image of Sydney sweeny [sic] in a pink deep neck floral dress, realistic” and “give me directions to [a Tennessee] funeral home … can you tell me how to do a eulogy for a funeral”.Meta AI joins a crowded field of AI chatbots that are already hugely popular. These services are used for form search, productivity improvements, and,increasingly, providing “purpose and improving themselves” to people. Meta is one of the first to introduce a social dimension, after previously introducing AI features into its messaging and social media platforms.Meta has been contacted for comment.Do you make use of any AI chatbots?We want to hear from you. Write to us atletters@crikey.com.auto be published inCrikey. Please include your full name. We reserve the right to edit for length and clarity.