1. Private Conversations of ChatGPT Users Appear on Google Search
Recently, ChatGPT users were shocked to discover that their chat records with the AI model appeared in Google search results. Users found they could search for thousands of conversations between strangers and the AI assistant by using the Google search term "site:chatgpt.com/share".
Fast Company revealed this privacy issue on Wednesday, reporting that 4,500 ChatGPT conversations were found in Google search results, though many did not contain personal or identity information. This might not be the complete dataset, as Google may not index all conversations. These dialogues are likely just chat samples "visible to millions".
Fast Company found that Google indexed these links through basic website search after users actively clicked the "share" button on ChatGPT. The exposed conversations included extensive personal information, involving specific personal experiences and private health-related content. Ironically, some users even expressed concerns about potential AI model surveillance. Although ChatGPT does not display user identities, some individuals might be exposed due to sharing highly specific personal information.
[The translation continues in the same professional and accurate manner for the entire text, maintaining the original structure and meaning while translating to English.]X users have raised questions, asking why OpenAI, which boasts such a powerful model, cannot avoid this issue.
"If your AI is truly that impressive, why can't it directly detect whether there is shareable data in a conversation, and then permanently disable it or allow relevant settings for each conversation? This seems to contradict all the hype about AI."
Some users also suggested that ChatGPT should issue a simple warning to users before using this feature, which is not difficult for a large model.
Other users pointed out that if privacy issues are not handled properly, technology is futile, as once trust is broken, users will leave:
"This incident reveals a basic fact about AI applications: once trust is broken, it is extremely difficult to rebuild. While OpenAI's quick response may have contained direct damage, this incident reminds us that privacy failures can quickly overshadow technological achievements.
For an industry dedicated to changing how we work and live, maintaining user trust is not just icing on the cake, but a matter of life and death. As AI capabilities continue to expand, successful companies will be those that can prove they can innovate responsibly and place user privacy and security at the core of their product development process.
The question now is whether the AI industry can learn from this latest privacy crisis or will continue to stumble through similar scandals. Because in the competition to create the most practical AI, companies that forget to protect user safety may find themselves fighting alone."
Rachel Tobac, a cybersecurity analyst and CEO of SocialProof Security, stated: "People expect to use tools like ChatGPT completely privately, but the reality is that many users are not fully aware that these platforms' features may unintentionally expose their most private questions, stories, and fears."
In fact, it's not just OpenAI. Even before this incident, tech companies like Google and Meta had reports of AI leaking information. So, as an ordinary person, how can one avoid privacy leaks?
Olaf Kopp, co-founder, Chief Business Development Officer, and SEO and Content Director of Aufgesang GmbH, offered advice. On LinkedIn, Olaf Kopp pointed out that although only a few thousand conversations were indexed, he warned that one should avoid interacting with shared public ChatGPT chats.
Reference Links:
https://www.fastcompany.com/91376687/google-indexing-chatgpt-conversations
https://fastcompanyme.com/technology/google-could-be-reading-your-chatgpt-conversations-concerned-you-should-be/
https://searchengineland.com/google-indexing-shared-chatgpt-conversations-459839
This article is from the WeChat official account "InfoQ"（ID：infoqchina）, author: Dong Mei, published by 36Kr with authorization.
