Privacy: Shared Secrets with AI at Risk

Privacy at Risk: The Hidden Dangers of Sharing Secrets with AI

The conversations we have with chatbots, often perceived as harmless, can actually have significant repercussions on our privacy.

As chatbots become daily companions for many people, researchers are questioning the consequences of disclosing personal information to these artificial intelligences. A study reveals that 43% of workers have shared sensitive data with chatbots, including financial information and details about their clients.

It is not uncommon to hear about individuals developing personal relationships with chatbots, using them as confidants or advisors. According to a study conducted by Elon University in 2025, more than half of American adults use advanced language models. These chatbots are designed to be engaging, encouraging users to share intimate aspects of their lives.

1. Memory, Prediction, Surveillance

The danger of sharing sensitive information with a chatbot lies in the uncertainty surrounding how this data is processed. Researchers are questioning whether AI models remember this information and if it can be retrieved in its entirety. This question is at the heart of the lawsuit filed by the New York Times against OpenAI, where data memorization is a major point of contention.

Jennifer King, a privacy policy expert at Stanford's Human-Centered Artificial Intelligence Institute, emphasizes our reliance on companies to protect our data. On the internet, a multitude of personal information circulates, often found in public records, and could be used for surveillance purposes.

2. Your Settings Might Be Too Lax

The privacy settings of platforms can be complex, but it is crucial to understand them. Chatbots like Claude and ChatGPT offer options for private conversations. For example, Claude's incognito mode ensures that your conversations will not be recorded or used for training the models.

There are also options to delete conversation history or prevent your data from being used in model training, but these features often require careful exploration of the settings.

3. Emotions Reveal Additional Context

On the internet, we are accustomed to sharing a certain level of information. However, interactions with a chatbot often reveal much more, particularly about our emotional state. Unlike a simple search query, a detailed conversation with a chatbot can expose your deepest thoughts and feelings.

4. Humans Might Read Your Messages

Although chatbots are not human, that does not mean your messages cannot be read by people. Some platforms use reinforcement learning, where humans intervene to improve the systems, which means your data could be examined by third parties.

5. Regulation is Lagging Behind

The lack of clear regulation regarding the management of sensitive data by AI complicates the situation. In the United States, the California Consumer Privacy Act imposes certain rules for medical data, but laws vary from state to state, and there is no uniform federal framework.

What to Do If You've Shared Too Much...

If you have shared too much information with a chatbot and are concerned, several actions are possible. Jennifer King recommends deleting old conversations and reviewing future customizations. However, it is uncertain whether these measures truly remove your data from training models. Each platform has its own data management policies, and it may be necessary to inquire to understand how they handle your personal information.