Mithril Security is a security startup focused on democratizing privacy-by-design AI through open-source projects.

Our main product is BlindChat, a privacy-first Conversational AI, that helps users leverage AI with guarantees their data always remains private.

Thanks to our confidential infrastructure, not even our admins are able to see the prompts sent to our Conversational AI as they remain end-to-end protected. This tackles the growing privacy concerns related to using Conversational AI by enabling users to query AI models without exposing their data to the AI provider.

BlindChat is deployed with BlindLlama, our open-source project to serve AI privately by leveraging enclaves.

BlindChat builds on our previous open-source projects (some of which are no longer maintained):

• BlindAI, a solution for querying and deploying AI models while guaranteeing end-user data privacy, audited by Quarkslab.
• BlindBox, a SaaS deployment solution that boosts compliance and improves the security posture of SaaS solutions by shielding end users' data at all times- including from the SaaS providers themselves.
• BastionLab, an access-control privacy framework for remote data exploration and AI training.

• Mithril Security's Blog - We regularly post articles and demos relating to privacy and security in AI and secure hardware.
• Confidential Computing Explained- This is our series of beginner-friendly tutorials covering how to implement confidential computing solutions.

You can check out our repositories and give us a ⭐️ if you like our projects!