Signal co-founder launches privacy-focused AI service Confer

In December, Moxie Marlinspike, the co-founder of the nonprofit organization Signal, launched Confer, a new privacy-conscious artificial intelligence service. The product is designed to look and feel like AI services such as ChatGPT, which is owned by OpenAI, or Claude. However, unlike those platforms, Confer’s back end is arranged specifically to avoid data collection. Because the host does not have access to user conversations, these interactions cannot be used to train the model or target advertisements. This architecture aims to provide a functional alternative to models that retain user data, allowing users to have sensitive conversations with the model without any information leaking out.

The launch of Confer is a response to the highly intimate nature of modern conversational AI. Marlinspike has warned that the technology actively invites users to share deeply personal details, describing it as a system that actively invites confession. He argues that the data collection models that fuel companies like Facebook and Google could creep into chatbot conversations, especially as companies like OpenAI test advertising on their platforms. “Chat interfaces like ChatGPT know more about people than any other technology before. When you combine that with advertising, it’s like someone paying your therapist to convince you to buy something,” said Marlinspike.

To deliver on its privacy promises, Confer relies on a technical stack that secures data both in transit and during processing. First, the service encrypts messages to and from the system using WebAuthn, which is a passkey authentication standard. On the server side, Confer’s inference processing—the process of running data through a trained AI model—is performed entirely within a Trusted Execution Environment (TEE), which is a secure area of a main processor. To ensure the system remains secure, Confer uses remote attestation, a process that verifies the integrity of the system and ensures it has not been compromised. Inside this secure environment, an array of open-weight foundation models—which are AI models with accessible weights—handles user queries.

Confer is available under two distinct tiers:

• A free tier limited to 20 messages a day and five active chats.
• A paid tier that costs $35 a month.

Why it matters

Confer represents a shift toward privacy-preserving AI architecture, utilizing hardware-level security to offer a functional alternative to mainstream models that rely on user data retention.