As AI assistants become woven into everyday life, concerns about privacy are growing harder to ignore. Using these tools almost always means sharing deeply personal information, which is typically stored and controlled by the companies behind the models. With OpenAI already experimenting with advertising, many users worry that chatbot conversations could eventually be mined in the same way social media platforms exploit personal data.
A new initiative offers a very different vision. In December, Signal co-founder Moxie Marlinspike introduced Confer, an AI assistant built from the ground up with privacy as its core principle. On the surface, Confer feels familiar—its interface closely resembles tools like ChatGPT or Claude. Behind the scenes, however, the architecture is radically different, designed to prevent data collection altogether. Conversations are never accessible to the service operator, which means they can’t be used for model training or targeted advertising.
Marlinspike argues that this level of protection is essential because of how people interact with AI chatbots. These systems naturally encourage users to open up, sharing thoughts and concerns they might never post publicly. In his view, combining that intimacy with advertising creates a serious ethical problem—comparable to paying a therapist to subtly influence a client’s purchasing decisions.
Delivering that promise of privacy requires multiple layers of security working together. Messages sent through Confer are encrypted end-to-end using the WebAuthn passkey system, reducing the risk of interception. While this standard works most smoothly on mobile devices and newer macOS versions, it can also be used on Windows or Linux with the help of a password manager.
On the backend, Confer processes all AI inference inside a Trusted Execution Environment (TEE). Remote attestation is used to verify that the system hasn’t been tampered with, ensuring that even the servers themselves cannot inspect user conversations. Within this secure environment, a selection of open-weight foundation models handles user requests without exposing the underlying data.
This setup is significantly more complex than a conventional AI inference pipeline, but it enables Confer to fulfill its central claim: users can discuss sensitive topics without worrying that their information will be stored, analyzed, or monetized.
Confer offers a free plan capped at 20 messages per day and up to five active conversations. A paid subscription costs $35 per month and unlocks unlimited usage, access to more advanced models, and additional personalization features. While the price is higher than ChatGPT Plus, Confer makes a clear tradeoff—prioritizing privacy over scale, and accepting that true data protection comes at a cost.
