Signal co-founder Moxie Marlinspike has unveiled Confer, a new privacy-focused AI service designed to offer the utility of chatbots like ChatGPT without the associated data surveillance. Launched in December, the project aims to counter the growing trend of AI models retaining personal information for training and advertising purposes.
Marlinspike argues that current AI assistants invite intimate disclosure, creating a dangerous dynamic when paired with the profit motives of big tech. “Chat interfaces like ChatGPT know more about people than any other technology before,” Marlinspike noted. “When you combine that with advertising, it’s like someone paying your therapist to convince you to buy something.”
To prevent this, Confer is built on a rigorous architecture that ensures the host never has access to user conversations. The system employs WebAuthn passkeys to encrypt messages on the user’s device and processes all inference within a Trusted Execution Environment (TEE) on the server side. Remote attestation systems verify that the server hasn’t been compromised, while open-weight foundation models handle the actual queries inside this secure enclave.
The service operates on a freemium model. A free tier restricts users to 20 messages a day and five active chats, while a $35 per month subscription offers unlimited access, advanced models, and personalization features. While significantly more expensive than competitors like ChatGPT Plus, Confer pitches the premium as the necessary cost of true privacy.
