The conversational nature of modern artificial intelligence tools makes it incredibly easy to let our guard down, but those casual chats might not be as private as you think. From casual queries to sensitive personal dilemmas, people are disclosing a vast amount of data to AI chatbots, potentially exposing themselves to unforeseen repercussions.

The Data Training “Black Box” One of the primary concerns surrounding AI privacy is how companies handle the data you input. According to Dr. Jennifer King of Stanford University, who led a study on AI and privacy, companies retain user chats to train and improve their AI agents. The major issue is a lack of transparency regarding this process; it remains unclear exactly what companies do with the data, what specific information they remove, or how they attempt to clean it before use. Dr. King notes that from the outside, these internal processes are largely a “black box”.

Furthermore, the legal landscape surrounding AI data is actively evolving. Users should be aware that chat logs are not immune to legal scrutiny. For instance, a user’s AI chat records could potentially be subpoenaed and used against them in a court of law, such as by an ex-partner during a dispute.

How Long Do Platforms Keep Your Data? A closer look at the policies of major AI platforms reveals varying approaches to data retention:

• ChatGPT: Conversations may be stored and used to improve models, though the platform states this is not done in a way that personally identifies the user. Chats are retained until the user deletes them, after which they are removed within 30 days. However, if you are using an enterprise version provided by your employer, they may hold onto the data for longer periods.
• Claude: Conversations are saved to the user’s account until deleted. If a user opts into data training, Anthropic can hold onto the data for up to five years to feed into its training cycles.
• Gemini: Chats are saved to your Google account to improve services. Users have the ability to view, delete, or completely turn off this activity at any time via their Google account settings.

It is also important to note that while all three platforms offer delete options, they state that if chats are flagged for any reason, the data may be held longer for legal purposes.

Expert Advice for Protecting Your Privacy Given the current uncertainties, privacy experts urge users to be highly judicious about what they choose to share with AI. Dr. King offers several actionable steps to protect your data:

• Redact Personal Information: Never upload files containing sensitive data—such as medical records for help with bills, or personal photos—without completely redacting all identifying information first.
• Use Temporary Chats: Opt for temporary chat features when available, as it is possible these are used less frequently for AI training data.
• Opt Out of Training: Dig into the settings of your chosen AI tool and explicitly opt out of allowing your information to be used for model training.

Ultimately, until there is more transparency in how these tools operate behind the scenes, exercising extreme caution is the best defense when navigating conversations with AI.