Introducing Confer: Moxie Marlinspike’s Privacy-Focused AI Chatbot
In an era where artificial intelligence (AI) personal assistants are becoming ubiquitous, concerns about user privacy are escalating. Engaging with these AI systems often necessitates sharing personal information, which is typically stored and potentially utilized by the companies behind these technologies. The recent move by OpenAI to experiment with advertising within its platforms has intensified apprehensions about data collection practices infiltrating AI interactions.
Addressing these privacy concerns, Moxie Marlinspike, co-founder of the encrypted messaging service Signal, introduced a new AI project in December: Confer. This platform mirrors the user experience of popular AI chatbots like ChatGPT and Claude but is architected with a strong emphasis on user privacy. Confer’s design ensures that user conversations are inaccessible to the host, thereby preventing their use in model training or targeted advertising.
Marlinspike underscores the significance of these privacy measures, highlighting the deeply personal nature of interactions with AI chat interfaces. He notes that such technologies actively invite confession, accumulating more personal data than any previous technology. Combining this data collection with advertising mechanisms, he argues, is akin to someone paying your therapist to convince you to buy something.
To uphold user privacy, Confer employs a multifaceted approach:
1. End-to-End Encryption: Confer utilizes the WebAuthn passkey system to encrypt messages between users and the platform. While this standard is optimized for mobile devices and Macs running Sequoia, it can also be configured on Windows or Linux systems using a password manager.
2. Secure Processing Environments: All inference processing within Confer is conducted in a Trusted Execution Environment (TEE). This setup includes remote attestation systems to verify the integrity of the environment, ensuring it remains uncompromised.
3. Open-Source Foundation Models: Confer operates using a suite of open-weight foundation models to handle user queries, promoting transparency and trust in the system’s operations.
While this infrastructure is more complex than standard AI inference setups, it effectively delivers on Confer’s promise of privacy. Users can engage in sensitive conversations with the assurance that their information remains confidential.
Confer offers a free tier that includes 20 messages per day and five active chats. For users seeking unlimited access, advanced models, and personalized features, a subscription is available at $35 per month. Although this pricing exceeds that of ChatGPT’s Plus plan, it reflects the premium placed on privacy and data security.
In summary, Confer represents a significant advancement in AI personal assistants by prioritizing user privacy. Marlinspike’s initiative offers a compelling alternative for individuals seeking the benefits of AI interactions without compromising their personal data.
