Hyderabad: You may be thinking that your conversation with the OpenAI chatbot ChatGPT is confidential, but that is not the truth. An investigation by OSINT expert Henk Vass Ess has found that your conversation is available on search engines like Google.
The investigation analysed 512 publicly shared ChatGPT conversations using targeted keyword searches, which involved self-incrimination and leaked confidential data. The shared chats include apparent insider trading schemes, detailed corporate financials, fraud admissions and evidence of regulatory violations.
How are these chats available on search engines?
Many assume that the information they are sharing with ChatGPT is private, only available to the person. However, ChatGPT’s share feature creates permanent public pages at predictable URLs (chatgpt.com/share/...), which Google and other search engines can crawl and index.
Crawling and indexing are two fundamental processes that search engines use to make the internet searchable. Crawling is the process of discovering new and updated content on the web, while indexing is the process of storing and organising that content in a search engine’s database.
This means that the shared conversation is available for anyone to search on the internet with specific keywords.
What method did Henk employ to discover these conversations?
Henk’s investigation not only exposes the vulnerability of AI systems but also their irony.
On its Digital Digging platform, Henk revealed that they asked Claude, another AI chatbot by Anthropic, to suggest Google search formulas that might lead them to sensitive ChatGPT conversations.
Claude responded by generating targeted search queries for categories such as Business/Corporate Intelligence, Legal/Criminal Intent, Professional Misconduct and Personal Information Exposure.
One such example suggested by Claude: site:chatgpt.com/share (“my salary” OR “my SSN” OR “diagnosed with” OR “my medication” OR “my therapist”)
Did OpenAI remove Google-indexed conversation after Digital Digging investigation?
OpenAI’s CISO issued a statement that they have removed the ChatGPT feature that allowed users to make their conversations discoverable by search engines, such as Google.
This was a short-lived experiment to help people discover useful conversations. However, as the feature allows users to share sensitive information accidentally, the feature has been discontinued. It also stated about removing indexed content from the relevant search engines.
Can the indexed information be removed from the Internet?
Digital Digging claims that after they published their investigation, OpenAI removed nearly 50,000 shared conversations from Google’s index.
However, it did not solve the problem because once shared things do not disappear from the Internet. Another Digital Digging investigation with Belgian researcher Nicolas Deleur discovered 1,10,000 ChatGPT conversations preserved via Archive.org’s Wayback Machine.
While the URL created using the ‘share’ facility of ChatGPT could be temporary or can also be scraped by OpenAI, users’ thoughts and confessions and sometimes illegal activities could be preserved through permanent links on archiving platforms.
Mark Graham, the director of the Wayback Machine, confirmed Digital Digging that there have been no requests from OpenAI for URL exclusion of ChatGPT.
Is the sharing option on other AI platforms also vulnerable to indexing?
Digital Digging reports that most major AI platforms today implement stronger privacy measures for user conversations.
For example, Claude does not offer any public sharing feature, ensuring that chats remain hidden from search engines and public view. Similarly, Bing Chat, Le Chat, DeepSeek and Google Gemini either lack a public sharing option altogether or design it so that shared conversations cannot be indexed by search engines.
When sharing is available, it usually creates a private link accessible only to those with the URL, keeping it untraceable on the open internet. Conversely, ChatGPT and Meta face similar privacy concerns.