The emergence of AI personal assistants has prompted privacy concerns regarding personal data retention and potential use for advertising by model providers.
In December, Signal co-founder Moxie Marlinspike initiated Confer, an AI service that prioritizes user privacy. Designed to operate similarly to other AI chatbots, Confer's backend architecture is structured to prevent data collection. User conversations with Confer are not utilized for model training or advertising purposes.
Marlinspike stated that AI chat interfaces invite personal disclosures, and combining this with advertising practices is comparable to a therapist being paid to influence a client's purchasing decisions.
Confer's privacy infrastructure involves several components. Messages are encrypted using the WebAuthn passkey system. On the server, all inference processing occurs within a Trusted Execution Environment (TEE), which incorporates remote attestation systems to verify its integrity. Within the TEE, open-weight foundation models handle user queries.
This system, while more complex than standard inference setups, is intended to ensure user privacy during sensitive interactions with the model.
Confer offers a free tier limited to 20 messages daily and five active chats. A subscription for $35 per month provides unlimited access, advanced models, and personalization features. The article notes this pricing is higher than some competing services, attributing the difference to the privacy features offered.