# How to Use AI With Your Privacy Intact > Most AI chatbots collect your deepest secrets by default. Here's how to talk to them safely. - Title: How to Use AI With Your Privacy Intact · WIRED | Shortify - Summary: Most AI chatbots collect your deepest secrets by default. Here's how to talk to them safely. Privacy-preserving AI exists but requires paying more, accepting… - Keywords: ai chatbots, privacy, data collection, encryption, chatgpt, claude, gemini, end-to-end encryption, trusted execution environments, zero data retention, signal, confer - Source: WIRED — https://wired.com/story/how-to-use-ai-with-your-privacy-intact - Author: Andy Greenberg - Published: 2026-09-22T09:00:00+00:00 - Read time: 3 min - Topics: ai chatbots, privacy, data collection, encryption, chatgpt, claude, gemini, end-to-end encryption, trusted execution environments, zero data retention, signal, confer ## AI chatbots collect your secrets ChatGPT, Claude, Gemini and similar tools store your conversations by default with few restrictions on how companies share, sell, or hand over that data to law enforcement. ## You're building a profile When you ask an AI chatbot personal questions about your life, finances, health, and relationships, you're creating a detailed profile that the company can access and use. > You're giving it this huge profile on you. ## Privacy shifted from texts to AI Signal creator Moxie Marlinspike says AI conversations now pose the same surveillance risks that unencrypted text messages did a decade ago, but far more severe. ## Confer offers encrypted AI Marlinspike launched Confer, an AI chatbot that uses cryptography to prevent even the service's own servers from reading or logging your conversations. > Confer is designed to be a service where you can explore ideas without your own thoughts potentially conspiring against you someday. ## New privacy-focused AI tools emerge A growing wave of AI services now promise stronger privacy protections through zero data retention policies, anonymization, or technical safeguards. ## Enterprise zero data retention OpenAI, Anthropic, and Google offer zero data retention policies for paid enterprise accounts that legally require them to delete conversations immediately after processing. ## Enterprise policies have limits Even zero data retention policies exclude some advanced models due to misuse concerns, and companies still monitor for abuse before deleting data. ## Most users get only promises Consumer AI services like Proton's Lumo and DuckDuckGo's Duck.ai promise not to log conversations, but these are policy pledges, not technical guarantees. > 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 offer partial privacy Duck.ai and Venice.ai relay your requests through their own servers to Claude or ChatGPT, limiting what those companies learn about you but not eliminating data collection. ## Your messages reveal your location Even anonymized messages can identify you. Asking about coffee shops in your neighborhood reveals where you live, creating a fingerprint from combined data. > All that stuff tied together is basically a fingerprint for you. ## Trusted execution environments work Some AI services use hardware-based trusted execution environments that cryptographically prevent servers from accessing your queries, even though they can't be fully end-to-end encrypted. ## Confer uses the strongest tech Confer combines Nvidia's confidential computing hardware with passkey authentication and open-source code to offer the strongest consumer-level AI privacy protections available. ## Meta AI in WhatsApp has privacy mode WhatsApp's Meta AI chatbot includes an incognito mode using trusted execution environments that prevents Meta from keeping records of those conversations. ## Apple's private cloud compute Apple Intelligence uses Private Cloud Compute, which cryptographically verifies that servers run unaltered code designed to keep no logs and isolate your data. ## Run AI locally for total privacy Tools like Ollama, LMStudio, and LocalAI run on your computer with zero data leaving your device, but they're far less capable than cloud-based models. > They're just dumb as rocks. ## Privacy costs more money Confer charges $34.99 monthly after free queries run out, more than ChatGPT or Claude, because it lacks data monetization subsidies. > It's more expensive, but it's the real cost. ## Privacy AI needs sustainable models Without profits from data collection or advertising, it's unclear whether privacy-focused AI companies can sustain themselves long-term without eventually compromising their promises. ## Real privacy requires payment Privacy-preserving AI exists but demands higher prices, weaker capabilities, or trusting company promises instead of technical guarantees. > If you want real privacy, you might have to pay for it. ## Key takeaway Privacy-preserving AI exists but requires paying more, accepting weaker models, or trusting company promises instead of technical guarantees.