How to Keep Your Private Conversations Safe From AI Companies

Modern AI chatbots have become intimate repositories of human secrets. Millions of people turn to tools like ChatGPT, Claude, and Gemini to discuss personal insecurities, financial worries, health concerns, and relationship struggles. Yet by default, most of these platforms store that sensitive data indefinitely and can share it with partners, law enforcement, or courts—a fact that has alarmed privacy advocates.

Moxie Marlinspike, the cryptographer who created Signal, sees troubling parallels to the surveillance risks he fought against in encrypted messaging. "People are integrating AI into their personal lives," he explains. "They talk with it about their deepest insecurities, their finances, their health, their relationships." This realization spurred him to launch Confer earlier this year, an AI chatbot designed to prevent even the service itself from accessing users' conversations through advanced encryption.

Yet Confer is just one of a growing wave of AI tools promising greater privacy. As AI becomes unavoidable, protecting your data now requires understanding the varying levels of protection available—and what each costs you.

Enterprise Zero-Data Retention: Privacy for Corporate Users#

The gold-standard privacy protection is zero data retention (ZDR): a contractual obligation that forces AI providers to delete conversations immediately after processing. OpenAI, Anthropic, and Google all offer ZDR for enterprise and developer accounts.

These policies come with meaningful caveats. Anthropic excludes its most advanced models from ZDR due to safety concerns about potential hacking or misuse. OpenAI analyzes user activity before deletion to flag abuse patterns, and Google still logs some prompts for safety monitoring—though stripped of identifying markers. Critically, ZDR policies remain unavailable to ordinary users; only corporate clients can demand this protection.

Consumer Promises: Trust-Based Privacy#

For everyone else, privacy depends largely on a company's word. Services like Proton's Lumo and DuckDuckGo's Duck.ai promise not to log conversations, relying on their reputations rather than technical guarantees. Proton co-founder Andy Yen acknowledges the limitation: "A promise is not as good as a mathematical guarantee, but a promise made by the right people is still actually quite substantial."

Proxy services like Duck.ai and Venice.ai add a layer of indirection by routing requests through third-party models (Claude, ChatGPT) without revealing your identity to those providers. However, this approach has a crucial weakness: your actual conversations can still be logged by the underlying AI model, and even metadata-stripped messages often contain identifying details. Asking about neighborhood coffee shops, for instance, reveals your location—data that combines into a fingerprint unique to you.