The proliferation of artificial intelligence personal assistants has raised significant privacy concerns among users, largely due to the inherent sharing of personal information with model parent companies. This data retention opens the door to potential advertising targeting and extensive data collection, mirroring practices seen with platforms like Facebook and Google. In response to these growing anxieties, Signal co-founder Moxie Marlinspike launched Confer in December, an innovative project designed to demonstrate what a truly privacy-conscious AI service could offer.

Confer is engineered to provide a user experience similar to popular AI chatbots such as ChatGPT or Claude, but with a fundamentally different backend architecture that prioritizes the avoidance of data collection. Drawing on the open-source rigor that has established Signal's reputation for trustworthiness, Confer explicitly guarantees that user conversations will not be utilized for model training or targeted advertising. This assurance stems from the core design principle that the host company will never gain access to the conversational data itself.

Marlinspike emphasizes that these robust privacy protections are a direct response to the deeply intimate nature of AI chat interfaces. He points out that technologies like ChatGPT 'actively invite confession' and gather more personal insights than any preceding technology. The combination of such intimate knowledge with advertising, Marlinspike argues, is akin to a therapist being paid to influence a patient's purchasing decisions, underscoring the critical need for a privacy-first approach.

Achieving this level of privacy necessitates the coordinated operation of several intricate systems. Firstly, Confer employs the WebAuthn passkey system to encrypt all messages exchanged with the service. While this standard is optimally supported on mobile devices and Macs running Sequoia, it can also be configured for Windows or Linux through the use of a password manager.

On the server side, all of Confer's inference processing is conducted within a Trusted Execution Environment (TEE). These TEEs are fortified with remote attestation systems, which continuously verify the integrity of the system to ensure it has not been compromised. Within this secure environment, an array of open-weight foundation models processes incoming queries.

This sophisticated setup is considerably more complex than a standard inference architecture, which is already a nuanced technological challenge. However, this complexity is essential to fulfill Confer's fundamental commitment to its users: enabling sensitive conversations with the AI model without any risk of information leakage. This diligent implementation ensures that user data remains private and secure.

Confer offers a free tier that permits up to 20 messages daily and maintains five active chats simultaneously. For users requiring more extensive access and advanced features, a premium subscription is available for $35 per month. This paid tier provides unlimited access, along with access to more sophisticated models and enhanced personalization capabilities. While this monthly fee is notably higher than alternatives like ChatGPT's Plus plan, it reflects the premium associated with uncompromising privacy in the AI landscape.