In an article in Ars Technica, security journalist Dan Goodin reports that Moxie Marlinspike, the creator of the Signal messaging app, is developing Confer, an AI assistant designed to bring end-to-end encryption to AI interactions, much the way Signal transformed private messaging.
The piece draws on additional coverage from TIME Magazine and Forbes, which explore how Confer’s architecture sets it apart from mainstream AI assistants currently on the market.
Below are key highlights from the reporting:
- Project & purpose: Marlinspike aims to give users the same privacy guarantees in AI conversations that Signal brought to text messaging, meaning only the user can read their prompts and the AI’s responses, with no exposure to third parties or even the service provider itself.
- Encryption & identity: Confer encrypts user prompts, AI responses, and all associated data on the client side. Like Signal, it avoids linking users to their real-world identity through email addresses, IP addresses, or other common identifiers.
- Cryptographic tools: The system uses passkeys to generate unique key pairs for each service, storing the private key only on the user’s device. This approach enables strong client-side encryption and forward secrecy, meaning past conversations remain protected even if a key is later compromised.
- Trusted Execution Environments (TEEs): On the server side, Confer uses TEEs, a technology that protects data and code inside the server’s CPU, preventing even administrators from accessing user content. The server also offers remote attestation and a transparency log, allowing independent observers to verify what software is actually running.
- Syncing across devices: Users can securely sync encrypted conversations across multiple devices because decryption only happens on authorized devices. Despite the complexity of the underlying cryptography, the interface is described as straightforward.
- Broader landscape: While other privacy-oriented AI products exist, the reporting notes that mainstream AI providers do not currently offer the same level of end-to-end encryption protections that Confer is built around.
Goodin and the broader coverage emphasize that while privacy-focused AI tools are emerging, Marlinspike’s approach is notable for applying rigorous, Signal-grade cryptographic principles to a space that has largely overlooked them.
Acknowledgement: The summary section of this post was generated using AI writing tools as part of my AI Explorations process. For more information, please refer to the original articles listed below.