Moxie Marlinspike launches privacy-first alternative to ChatGPT

As artificial intelligence personal assistants become more widespread, growing concerns around privacy and data retention are prompting scrutiny over how much personal information users are required to share, particularly as AI companies explore advertising-led business models similar to those used by social media and search platforms.

A new project launched in December by Signal co-founder Moxie Marlinspike aims to address those concerns by offering a privacy-focused alternative to mainstream AI chatbots such as ChatGPT and Claude. The service, called Confer, is designed to resemble existing conversational AI tools in its interface and functionality, while fundamentally restructuring the backend to minimise data collection.

Marlinspike stated that Confer is built so that user conversations cannot be accessed by the host, used to train models, or leveraged for advertising. The system is designed with open-source principles similar to those used by Signal, ensuring that sensitive user interactions remain private.

He informed that chat-based AI services actively encourage users to share deeply personal information, noting that conversational interfaces may accumulate more intimate data than any prior consumer technology. He added that combining such data with advertising would introduce significant ethical concerns.

To enforce these privacy guarantees, Confer relies on multiple technical safeguards working together. Messages sent to and from the system are encrypted using the WebAuthn passkey system, which functions most effectively on mobile devices and Apple computers running macOS Sequoia, though it can also be configured on Windows or Linux using password managers.

On the server side, all inference processing takes place within a Trusted Execution Environment, with remote attestation systems in place to verify that the environment has not been compromised. Within this secure layer, an array of open-weight foundation models processes user queries without exposing the data externally.

While this architecture is significantly more complex than conventional AI inference systems, Marlinspike stated that it is necessary to fulfil Confer’s core promise of privacy, allowing users to engage in sensitive conversations without the risk of data leakage.

Confer currently offers a free tier limited to 20 messages per day and five active conversations. A paid subscription priced at $35 per month provides unlimited access, additional personalisation, and more advanced models, positioning it at a higher price point than OpenAI’s ChatGPT Plus plan.