WIRED: default AI chatbot privacy is near zero, guide warns

The starting position WIRED offers is blunt: with mainstream AI chatbots, users should assume "a baseline expectation of approximately zero real privacy from anyone who is determined to access your conversation records and has a legal path to obtaining them." That set includes the vendor, ad partners, contractors who help fine-tune the systems, law enforcement, and anyone who can subpoena records in a civil lawsuit.

Cryptographer Matt Green of Johns Hopkins puts the cost plainly. "You're giving it this huge profile on you," he tells Wired, and later: "If you want real privacy, you might have to pay for it." Moxie Marlinspike, Signal's creator, frames the stakes the way a chatbot user actually experiences them. "People are integrating AI into their personal lives. They talk with it about their deepest insecurities, their finances, their health, their relationships."

The piece then walks a ladder. Claude's conversations are not used for training by default; ChatGPT and Gemini require opt-out toggles ("Improve the model for everyone" off in ChatGPT, the equivalent training toggle off in Claude, the Keep activity setting off in Gemini). Paying $20 per month for ChatGPT Plus or Claude Pro does not change any of that. For stronger guarantees, enterprise customers can sign zero data retention (ZDR) contracts with OpenAI, Anthropic or Google that delete conversation records after processing. Above that sit no-log services like Proton Lumo, Venice.ai and Duck.ai that rely on policy rather than proof; trusted execution environment products such as Marlinspike's own Confer at $34.99 a month, Meta AI Incognito mode in WhatsApp, and Apple's Private Cloud Compute; and self-hosted local models like Ollama, LMStudio and LocalAI, which trade capability for privacy.