Go Ahead, Tell It Everything: Finally, an AI That Takes Your Secrets to the Grave

A new AI chatbot has emerged, promising to safeguard your most sensitive conversations. Confer, a generative AI assistant launched in December by Signal co-founder Moxie Marlinspike, aims to offer the convenience of tools like ChatGPT while ensuring that nothing you discuss is ever stored, analysed or monetised by its provider.

Unlike mainstream AI services, which retain user inputs to improve models or serve targeted advertising, Confer's architecture is built from the ground up to keep conversations confidential — even from the operators themselves. For users weary of sharing personal data with big tech, this represents a bold alternative in the field of artificial intelligence.

Read More: Leaked Signal Chat Shows NH Education Chair Kristin Noble Backing 'Whites-Only' Schools

Read More: Launch of 'Unregulated' ChatGPT Health in Australia: 'Misguidance or Harmful Misinformation?'

A Privacy-First Vision

Marlinspike, renowned for creating the encrypted messaging app Signal, says his latest project reflects deep unease about how conventional AI assistants treat user data. 'It's a form of technology that actively invites confession,' he told TechCrunch, arguing that chatbot interfaces often know more about users than ever before.

When coupled with advertising incentives, this could amount to 'someone paying your therapist to convince you to buy something,' he said.

To counter this, Confer eliminates standard data collection practices. Conversations are neither stored nor accessible to the hosting company, meaning they cannot be used to train the model or for targeted ads. This model of 'privacy by design' sets Confer apart from many popular AI services that treat user chats as valuable training data.

How It Keeps Secrets Safe

Confer's privacy guarantees rest on a sophisticated combination of encryption and secure computing. Messages between the user and the AI are encrypted using the WebAuthn passkey system, a modern authentication protocol designed to protect communications.

On the server side, all processing occurs inside a Trusted Execution Environment (TEE) — a hardware-isolated area that prevents even administrators from seeing the data being handled.

Remote attestation systems verify the integrity of these secure environments, ensuring that code hasn't been tampered with and that user data remains encrypted throughout. This approach reflects the high standards of confidentiality that underpinned Signal's own reputation in secure messaging.

Pricing and Accessibility

Confer operates on a tiered model. A free plan permits up to 20 messages per day and five active chats, suitable for casual users or those simply curious about private AI. For more frequent use, including unlimited messaging, personalised settings and access to advanced models, a subscription costs around $35 (£26.12) a month. That price is higher than similar tiers from major AI providers but reflects the complexity and cost of maintaining rigorous privacy protections.

While the WebAuthn passkey system works most smoothly on mobile devices and Apple's macOS Sequoia, it can also be adapted for Windows and Linux with a password manager, offering broad accessibility for diverse device users.

Privacy in an Age of Data Harvesting

Confer's arrival comes at a time when privacy issues are increasingly at the forefront of users' minds. OpenAI, for instance, has recently begun testing targeted ads in ChatGPT, sparking debate over data use in conversational AI.

For advocates of digital privacy, Confer may signal a turning point: an AI you can trust with your secrets, safe from prying eyes or corporate interests. Whether it will attract a broader audience remains to be seen, but it undoubtedly challenges the assumption that AI must come at the expense of privacy.