Moxie Marlinspike, the engineer behind the encrypted messaging application Signal, is now turning his attention to artificial intelligence with a new project called Confer, an open-source AI assistant designed to protect user data. Confer aims to provide assurances that user data is unreadable to the platform operator, hackers, law enforcement, or any party other than the account holders.

The service, including its large language models (LLMs) and back-end components, operates entirely on open-source software, allowing users to cryptographically verify its integrity. Data and conversations originating from users, along with the LLM's responses, are encrypted within a trusted execution environment (TEE). This encryption prevents even server administrators from accessing or tampering with the data. Confer stores conversations in the same encrypted form, utilizing a key that remains securely on the user's device.

The underlying mechanics of Confer are designed for both elegance and simplicity, similar to Signal's architecture. Signal established a new standard for end-user privacy tools by making strong encryption accessible and user-friendly. Marlinspike's goal with Confer is to replicate this success in the realm of AI, where concerns about data privacy and security are growing.

The development of Confer comes at a time when AI chatbots are becoming increasingly integrated into daily life. These AI systems often collect and process vast amounts of user data, raising concerns about potential misuse or unauthorized access. Encryption within a TEE is a security measure that creates an isolated and secure environment for processing sensitive data. This ensures that even if the main system is compromised, the data within the TEE remains protected.

The implications of Confer's approach extend beyond individual privacy. By ensuring that AI systems cannot access or misuse user data, Confer could foster greater trust in AI technology and encourage wider adoption. This is particularly important in sensitive areas such as healthcare, finance, and legal services, where data privacy is paramount.

The current status of Confer is in development, with Marlinspike and his team working to refine the system and expand its capabilities. The open-source nature of the project allows for community contributions and independent verification of its security claims. Future developments may include integrations with other applications and services, as well as enhancements to the AI models themselves.