No matter which chatbot you choose: the more you share, the more useful they become. The patient uploads blood report analysis, and the engineer pastes the unreleased code for debugging. But artificial intelligence experts remind us that we should be wary of such anthropomorphic tools, especially sensitive information such as social security numbers and corporate confidential data.
Although technology companies are eager to use data to optimize models, they do not want to obtain private user data. OpenAI clearly warns, "Do not disclose any sensitive information in the conversation." Google also emphasized to Gemini users: "Avoid entering confidential information or content that you do not want the reviewer to see."
Artificial intelligence researchers believe that chats about strange skin rashes or financial missteps could be used to train new versions of artificial intelligence or exposed through data breaches. The following lists what users should avoid typing and shares some privacy-protecting conversation tips:
Chatbots’ realistic, human expressions often put people off their guard. Jennifer King, a researcher at the Stanford Institute for Human-Centered Artificial Intelligence, cautions that once you enter information into a chatbot, “you lose control of that information.”
In March 2023, a ChatGPT vulnerability allowed some users to peek into the content of others' initial conversations. Fortunately, the company quickly fixed the vulnerability. OpenAI has also mistakenly sent subscription confirmation emails before, resulting in the leakage of user names, email addresses and payment information.
If you encounter a data hacker or need to cooperate with a judicial search warrant, chat records may be included in the leaked data. It is recommended to set a strong password for your account and enable multi-factor authentication, and avoid entering the following specific information:
Identification information.Include social security number, driver's license number and passport number, date of birth, address and contact number. Some chatbots have the function of automatically desensitizing such information, and the system will actively block sensitive fields. An OpenAI spokesperson said, "We are committed to letting artificial intelligence models learn about the world, not personal privacy, and actively reduce the collection of personal information."
Medical test report.Medical confidentiality is designed to prevent discrimination and embarrassment, but chatbots are generally not subject to special protections for medical data. If artificial intelligence is needed to interpret test reports, Jin from the Stanford Research Center recommends: "Please crop and edit the pictures or documents first, try to keep only the test results, and desensitize other information."
Financial account information.Bank and investment account numbers may become a breakthrough point for fund monitoring or theft, so they must be strictly protected.
Confidential Company Information. Users who prefer to use various generic versions of chatbots at work may inadvertently reveal customer data or trade secrets, even if they are just composing a common email. Samsung once banned the service completely after engineers leaked internal source code to ChatGPT. If artificial intelligence is really helpful for work, companies should choose a commercial version or deploy a customized artificial intelligence system.
Login credentials.With the rise of agents capable of performing a variety of real-world tasks, the need to provide account credentials to chatbots has increased. However, these services are not built according to digital vault standards, and passwords, PIN codes and security questions should be kept by professional password managers.
When users give positive or negative comments to robot responses, it may be deemed that questions and AI responses are allowed to be used for evaluation and model training. If a conversation is flagged for sensitive content such as violence, it may even be manually reviewed by company employees.
Claude, developed by Anthropic, does not use user conversations to train artificial intelligence by default and will delete the data after two years. Although OpenAI's ChatGPT, Microsoft's Copilot and Google's Gemini use conversation data, they all provide the option to turn off this feature in the settings. If users pay attention to privacy, you can refer to the following suggestions:
Delete records regularly.Jason Clinton, chief information security officer at Anthropic, advises cautious users to clean up their conversations promptly. AI companies typically purge data marked "deleted" after 30 days. Enable ad hoc conversations. ChatGPT's "temp chat" feature is similar to a browser's private browsing mode. When enabled, the user can prevent information from being stored in the user's profile. Such conversations are neither historically saved nor used to train the model.
Ask anonymously.The private search engine supports anonymous access to mainstream artificial intelligence models such as Claude and GPT, and promises that these data will not be used for model training. Although advanced operations such as file analysis of a full-featured chatbot cannot be implemented, basic Q&A is sufficient.
Keep in mind that chatbots are always happy to continue a conversation, but it’s always up to the user when to end the conversation or hit the “delete” button.