Moxie Marlinspike’s Privacy-First ChatGPT Alternative
▼ Summary
– AI personal assistants raise privacy concerns as they require sharing personal data, which companies retain and could potentially use for advertising.
– Confer is a new privacy-focused AI service designed to prevent data collection, ensuring conversations are not used for model training or ad targeting.
– The service uses encryption via WebAuthn and processes data in a secure Trusted Execution Environment to protect user information.
– Confer’s founder highlights the intimate nature of AI chat interfaces, warning that combining them with advertising exploits user trust.
– The service offers a free tier with limited messages and a paid plan at $35/month for unlimited access, prioritizing privacy over cost.
The growing use of AI assistants raises significant privacy concerns for many users, as personal conversations are often stored and analyzed by the companies behind the technology. This data collection, similar to practices seen in major advertising platforms, could potentially be used for targeted marketing within chatbot interactions. A new service called Confer, launched by Signal co-founder Moxie Marlinspike, presents a compelling alternative designed from the ground up to protect user privacy.
Confer offers a familiar chat interface comparable to ChatGPT or Claude, but its underlying architecture is fundamentally different. The system is built with open-source rigor and is structured to prevent data collection entirely. Your conversations cannot be used to train AI models or serve advertisements because the service host never has access to the content. For Marlinspike, this level of protection is a direct response to the uniquely intimate nature of AI chat. He describes these interfaces as a technology that “actively invites confession,” noting that they often learn more about individuals than any previous tool. Combining such deep personal insight with advertising models creates a troubling dynamic, akin to a therapist being paid to influence a client’s purchases.
Achieving this privacy standard requires multiple sophisticated systems working together. First, Confer encrypts all messages using the WebAuthn passkey system, which provides strong security, though it currently functions most seamlessly on mobile devices and newer Macs. On the server side, the platform performs all its AI processing within a Trusted Execution Environment (TEE), a secure, isolated area of a processor. This environment utilizes remote attestation to continuously verify that the system remains uncompromised. Inside this secured space, an array of open-weight foundation models handle user queries.
This intricate setup is far more complex than a standard AI inference pipeline, but it fulfills Confer’s core promise. As long as these protective measures remain intact, users can engage in sensitive discussions with the AI without fear of their information leaking. The service currently offers a free tier limited to 20 messages per day and five active chats. A subscription plan, priced at $35 per month, provides unlimited access, more advanced models, and personalization features. While this cost is notably higher than many mainstream AI subscriptions, it reflects the substantial investment required to build and maintain a truly private alternative.
(Source: TechCrunch)
