Moxie Marlinspike launches Confer, privacy-first AI chatbot with end-to-end encryption
In brief
- Moxie Marlinspike launched Confer in December 2025, a privacy-first AI chatbot with end-to-end encryption.
- Confer encrypts conversations on users' devices via WebAuthn passkeys and runs inference in Trusted Execution Environments.
- Architecture prevents Confer's operator from accessing, storing, or monetizing user conversations by design.
- Free tier offers 20 messages/day across 5 chats; premium subscription costs $35/month for unlimited access.
- Meta AI will integrate Confer's privacy framework by March 2026 across WhatsApp, Instagram, and Facebook.
Privacy by architecture
Confer uses WebAuthn passkeys to encrypt conversations on the user's device before transmission. On the server side, inference runs inside Trusted Execution Environments, or TEEs, which are isolated hardware enclaves that process data without exposing it to the underlying system or operators. Remote attestation allows users to cryptographically verify that the server is running what it claims to be running.
The result matters. Confer's operator cannot access, store, or monetize user conversations, not because of a policy that could be changed, but because of how the system is built. The entire technical framework is open-source and independently verifiable, meaning security researchers can audit the code themselves.
Industry contrast
The mainstream AI industry's standard practice is to process conversations in plaintext on centralized servers. That enables model improvement through user data and creates detailed behavioral profiles that are valuable to advertisers and governments. Confer inverts this model entirely.
Confer is available at confer.to and operates on a freemium model. The free tier allows up to 20 messages per day across five active chats, while a premium subscription runs $35 a month and unlocks unlimited access.
Meta integration ahead
The bigger picture emerges in March 2026. Confer's underlying privacy architecture is set to be integrated directly into Meta AI, which is embedded across WhatsApp, Instagram, and Facebook, platforms with billions of active users. That scale could reshape how AI conversations are handled at the infrastructure level.
Notably, Marlinspike's initiative for Confer deliberately eschews any integration with cryptocurrencies or blockchain technology, focusing on privacy engineering within the AI framework itself. The technical bet is clear: encryption and hardware isolation, not distributed ledgers, are the foundation for private AI.
Frequently asked questions
How does Confer encrypt conversations differently from standard AI chatbots?
Confer encrypts conversations on the user's device before transmission using WebAuthn passkeys, and processes queries in Trusted Execution Environments (TEEs), isolated hardware enclaves. Most AI platforms process conversations in plaintext on centralized servers, enabling data collection and behavioral profiling.
Why can't Confer's operator access user conversations?
The system's architecture prevents access by design, not by policy. Encrypted data on the user's device and inference running inside isolated hardware enclaves mean the operator has no technical means to access, store, or monetize conversations, even if they wanted to.
What's Confer's pricing and availability?
Confer is available at confer.to on a freemium model. The free tier allows 20 messages per day across five active chats. Premium costs $35 per month and unlocks unlimited access.
When will Confer's privacy tech reach billions of users?
Meta AI will integrate Confer's privacy architecture by March 2026. Meta AI is embedded across WhatsApp, Instagram, and Facebook, which have billions of active users combined.
