For those prioritizing privacy, the proliferation of AI personal assistants can be unsettling. Utilizing these services often necessitates sharing personal data, which is subsequently held by the parent company of the AI model. Given OpenAI’s ongoing ad trials, it’s conceivable that the extensive data gathering practices seen with Facebook and Google could soon extend to your chatbot interactions.
Moxie Marlinspike, co-founder of Signal, introduced a new initiative in December demonstrating how a privacy-focused AI service can operate. Named Confer, it offers an experience akin to ChatGPT or Claude, yet its underlying architecture is built to prevent data collection, employing the same robust open-source principles that establish Signal’s reputation. Your discussions on Confer are protected from being used for model training or targeted advertisements, precisely because the service provider will never obtain them.
Marlinspike views these safeguards as essential due to the deeply personal nature of such services.
Marlinspike states, “This technology inherently encourages users to confide personal details.” He adds, “Chat platforms such as ChatGPT accumulate more personal insights than any prior technology. Integrating this with advertising is akin to a third party compensating your therapist to influence your purchasing decisions.”
Achieving this level of privacy necessitates the coordinated operation of multiple distinct systems.
Firstly, Confer utilizes the WebAuthn passkey system to encrypt all communications with the platform. (This standard performs optimally on mobile devices or Macs utilizing Sequoia, though it can also function on Windows or Linux with a password manager.) Secondly, on the server, all of Confer’s inference tasks are executed within a Trusted Execution Environment (TEE), which incorporates remote attestation mechanisms to confirm the system’s integrity. Within this environment, various open-weight foundation models process incoming queries.
While this setup is considerably more intricate than a conventional inference architecture (which is already complex), it successfully fulfills Confer’s core commitment to its users. With these safeguards active, individuals can engage in private discussions with the AI without concern for data breaches.
Confer offers a free tier, restricted to 20 messages daily and five concurrent chats. Subscribers paying $35 monthly will receive unrestricted access, alongside enhanced models and personalization options. This price point is notably higher than ChatGPT’s Plus subscription, illustrating that robust privacy solutions often incur a greater cost.