AI chatbots now claim broader rights to your personal data

Artificial intelligence companies are expanding their claims to user data far beyond the chat window, creating new questions about how the intimate details people share with AI assistants might be used.

OpenAI recently introduced an optional Computer History feature that allows ChatGPT to track which apps and websites a person uses. Google announced it will use photos and other content uploaded through Search to train AI systems by default, though users can opt out. These moves signal a shift toward AI systems that maintain continuous awareness of users' digital lives.

Why it matters

Chatbots invite a different kind of disclosure than search engines or social networks. People ask AI assistants about health concerns, financial worries, workplace conflicts, and relationship problems—creating an unusually intimate record of their lives. As Meta, Google, and OpenAI all explore advertising models for their chatbots, that data could become increasingly valuable for targeting and persuasion.

The privacy spectrum

Company approaches to user data vary dramatically. Apple offers the most restrictive model with Apple Intelligence, processing most requests on-device and using Private Cloud Compute for complex tasks. Apple says it uses data only to fulfill requests and does not retain access to that content. However, this approach limits how much user history Apple can maintain for ongoing personalization.

Meta stakes a far broader claim in its privacy policy, stating it can use Meta AI interactions to personalize content and ads across its services, including conversations conducted through its smart glasses. The company says it excludes certain sensitive topics—health, politics, and religion—from ad personalization. Meta has also introduced an "Incognito Chat" mode for temporary conversations that are not stored.

Other services fall between these poles. Some offer temporary chat modes that bypass memory and training entirely. Others train on user data by default but allow opt-outs. Many let users view stored data and delete specific memories, though the mechanisms for doing so vary in complexity.

The new data bargain

"The AI era will increasingly be fueled by people voluntarily handing over their full digital lives to AI tools that promise to relieve their mental load or loneliness," Miranda Bogen, chief technologist at the Center for Democracy & Technology, told Axios. "The more a system knows about you, the easier it will be to make escalating requests for private details in a way that feels natural."

The critical question is no longer simply whether companies train AI models on user prompts. It is whether those prompts shape the system's understanding of individual users—and what else that understanding enables. A chatbot designed to keep users engaged longer, or to persuade them toward particular purchases or viewpoints, could draw on fears, insecurities, and private details shared in conversation.

Without comprehensive privacy regulations governing these scenarios, company policies carry unusual weight. Users who find value in personalized AI may accept the privacy tradeoff, but they need clarity about what they are agreeing to before they begin sharing.

These details were first reported by Axios as part of its ongoing series examining what AI companies know about their users.