Moxie Marlinspike (the cryptographic engineer behind the Signal messenger) has launched Confer, an AI assistant designed to solve the privacy nightmare of modern chatbots. Unlike standard platforms where providers can be legally compelled to turn over logs, Confer ensures that data remains unreadable to everyone except the user. The system treats AI interactions as private journals rather than public search queries, preventing prompts from being harvested into massive data lakes for training or monetization. The technical foundation of Confer relies on device-stored passkeys and hardware-level isolation. When a user interacts with the system, their device uses a unique private key to encrypt all inputs and outputs. This implementation provides forward secrecy, meaning that even if a specific session key is compromised, past and future conversations remain protected. To secure the server-side processing, Confer runs its Language Model inside a Trusted Execution Environment (TEE). A TEE functions as a secure black box on the server CPU that encrypts data while it is being processed, preventing system administrators from peeking at sensitive information. The platform also utilizes remote attestation, a cryptographic certificate that proves the server is running the exact, untampered open-source code it claims to be. This allows users to verify the privacy of the system mathematically rather than relying on corporate promises or vague terms of service. Confer aims for the same seamless simplicity that made Signal a global standard, currently providing native support for macOS, iOS, and Android. By combining cryptographic verification with user-friendly design, the project seeks to establish a new private AI infrastructure. This allows individuals to share sensitive business or personal thoughts with an assistant without the looming threat of data breaches, hacker theft, or government subpoenas.

Moxie Marlinspike (the cryptographic engineer behind the Signal messenger) has launched Confer, an AI assistant designed to solve the privacy nightmare of modern chatbots. Unlike standard platforms where providers can be legally compelled to turn over logs, Confer ensures that data remains unreadable to everyone except the user. The system treats AI interactions as private journals rather than public search queries, preventing prompts from being harvested into massive data lakes for training or monetization. The technical foundation of Confer relies on device-stored passkeys and hardware-level isolation. When a user interacts with the system, their device uses a unique private key to encrypt all inputs and outputs. This implementation provides forward secrecy, meaning that even if a specific session key is compromised, past and future conversations remain protected. To secure the server-side processing, Confer runs its Language Model inside a Trusted Execution Environment (TEE). A TEE functions as a secure black box on the server CPU that encrypts data while it is being processed, preventing system administrators from peeking at sensitive information. The platform also utilizes remote attestation, a cryptographic certificate that proves the server is running the exact, untampered open-source code it claims to be. This allows users to verify the privacy of the system mathematically rather than relying on corporate promises or vague terms of service. Confer aims for the same seamless simplicity that made Signal a global standard, currently providing native support for macOS, iOS, and Android. By combining cryptographic verification with user-friendly design, the project seeks to establish a new private AI infrastructure. This allows individuals to share sensitive business or personal thoughts with an assistant without the looming threat of data breaches, hacker theft, or government subpoenas.