Thousands of ChatGPT chats exposed on Google, what should users pay attention to?

From the incident of thousands of user conversations on ChatGPT being exposed on Google, security experts warn: AI is a useful support tool, but not a 'safe black box'.

Chat sharing feature via public link on ChatGPT - Photo: Techradar

ChatGPT now reaches over 2.5 billion queries from global users every day, showing the growing popularity of this chatbot.

However, this tool is facing strong backlash from users, as the “sharing” feature has caused thousands of conversations to leak sensitive information to Google and several search engines on the Internet.

High security risk

Mr. Vu Ngoc Son - head of technology of the National Cyber Security Association (NCA) - assessed that the above incident shows a high level of security risk for ChatGPT users.

“The above incident is not necessarily a technical flaw as there is a certain initiative of the user in clicking the share button. However, it can be said that the problem lies in the design of the AI chatbot product, when it has confused users and there are no strong enough warning measures about the risk of personal data being leaked if users share,” Mr. Vu Ngoc Son analyzed.

On ChatGPT, this feature is implemented after the user chooses to share the chat via a public link, the content will be stored on OpenAI's server as a public website (chatgpt.com/share/...), no login or password required to access.

Google's crawlers automatically scan and index these pages, making them appear in search results, including text, images, or sensitive chat data.

Many users were unaware of the risk, thinking they were sharing chats with friends or contacts. This led to thousands of conversations being exposed, some containing sensitive personal information.

Although OpenAI quickly removed the feature in late July 2025 after a backlash from the community, it still took time to coordinate with Google to remove the old indexes. Especially with the complex storage system, which includes Google's cache servers, this cannot be done quickly.

Don't treat AI chatbots as a "safety black box"

Data security expert Vu Ngoc Son - head of technology at the National Cyber Security Association (NCA) - Photo: CHI HIEU

The leak of thousands of chats can pose risks to users such as revealing personal secrets, business secrets; damage to reputation, financial risks or even safety hazards when revealing home addresses.

“AI-powered chatbots are useful but not a ‘black box’, as shared data can live on the web forever if left unchecked.

The incident is certainly a lesson for both providers and users. Other AI service providers can learn from this experience and design features with clearer, more transparent warnings.

At the same time, users also need to proactively limit the uncontrolled posting of personally identifiable or confidential information on AI platforms," security expert Vu Ngoc Son recommended.

According to data security experts, the above incident shows that there is a need for legal corridors and cybersecurity standards for AI. AI suppliers and developers also need to design systems that ensure security, avoiding the risk of data leakage such as: leakage through bad vulnerabilities; software vulnerabilities leading to database attacks; poor control leading to poisoning, abuse to answer misleading and distorted questions.

Users also need to control the sharing of personal information with AI, and not share sensitive information. In cases where it is really necessary, it is advisable to use anonymous mode or actively encrypt information to avoid data being directly linked to specific individuals.

Back to topic

PIETY

Source: https://tuoitre.vn/hang-ngan-cuoc-tro-chuyen-voi-chatgpt-bi-lo-tren-google-nguoi-dung-luu-y-gi-20250805152646255.htm