Mithril Security

What Mithril Security does

Mithril Security is a Paris-based open-source software company founded in April 2021 by Daniel Huynh (CEO), Raphaël Millet (COO), and Mehdi Bessaa (CTO). It builds tooling for confidential and trustworthy AI, with a portfolio centered on BlindLlama — an open-source framework that serves large language models inside hardware-based secure enclaves (Intel SGX, AMD SEV-SNP, vTPM) so that user prompts and model weights remain inaccessible even to the AI provider's administrators. Around this core, the company offers AICert, which creates cryptographic model provenance certificates binding model weights to training data and code; AIGovTool, an open-source governance framework for rate limiting, consumption control, and off-switch triggers on deployed AI models; and BlindChat, an end-user conversational interface with local and enclave modes. An earlier product, BlindAI (Rust-based ONNX inference on Intel SGX), was independently audited by Quarkslab with no vulnerabilities found but is no longer actively maintained.

The company's technology stack leverages trusted execution environments, TPM-based attestation, and Attested TLS to provide verifiable confidentiality and code integrity. Its go-to-market is community-led, with all products distributed freely on GitHub and supported by a Discord developer community, technical blog, whitepapers, and event participation. Enterprise adoption is driven through Calendly-based one-on-one consultations targeting healthcare, financial services, AI providers, and security/critical-infrastructure buyers with sensitive data or regulated workloads. Revenue appears to be generated via enterprise consulting, custom deployments, and likely recurring support or subscription engagements, though no public pricing is disclosed and revenue has not been reported.

Mithril Security has raised a 2022 pre-seed and a €1.2 million seed round in May 2023 led by multiple venture funds including Auriga Cyber Ventures, IT-Farm, Polytechnique Ventures, and SkyDeck Berkeley. It received an OpenAI Cybersecurity Grant in December 2023 and entered a flagship technical collaboration with the Future of Life Institute in November 2023 to develop hardware-backed AI governance tools. The company has stated intentions to expand into North American and European markets beyond its Paris headquarters.

Name

Mithril Security

Legal name

Mithril Security

Website

https://mithrilsecurity.io

Company type

Private

Founded year

2021

Operating status

Operating

Headcount range

11–50 employees

Short description

Mithril Security builds open-source tools for confidential AI inference and model provenance using secure enclaves and TPM-based attestation. It serves healthcare, finance, AI providers, and security-critical enterprises needing privacy-preserving LLM deployment.

Ownership category

akta.pro rank

Product category

Confidential AI infrastructure

NAICS

Software Publishers (5132), Computer Systems Design and Related Services (54151), Other Computer Related Services (541519)

SIC

Services-Prepackaged Software (7372), Services-Computer Programming Services (7371), Services-Computer Integrated Systems Design (7373)

akta.pro primary industry

Secure Model Deployment & Runtime Protection (sandboxing, isolation) (HDAAAKAH)

akta.pro secondary industries

Privacy-Preserving On-Device ML (secure enclaves, TEEs, differential privacy) (HDAAAJAH), Responsible AI, Security & Privacy Platforms (Safety, Guardrails, PII) (HDAEANAG), AI Privacy Engineering & Data Governance (PII, consent, retention) (HDAAAKAB)

Keywords

• Confidential AI inference
• Secure enclaves deployment
• AI model governance
• AI model provenance
• Trusted platform modules

Headquarters

HQ city

Ris-orangis

HQ country

France

HQ region

Europe

Offices1 record

Markets served

GTM type

B2B

Offering type

Software

Cost components

Technology or R&D, Personnel, Infrastructure, Marketing or Sales, Operations

Revenue model

1. Open Source Products: Open-source products freely available. Revenue likely generated through enterprise consulting, support services, and custom implementation engagements. Contact sales for tailored solutions.

Go-to-market motion2 records

Distribution channels2 records

Marketing channels6 records

Core offering

Mithril Security develops open-source software tools that enable AI providers to deploy models using hardware-based confidential computing. Its flagship product BlindLlama serves LLMs inside secure enclaves (Intel SGX, AMD SEV-SNP) with TPM-rooted attestation so user data is never accessible to AI provider administrators. The portfolio is extended with AICert for cryptographic AI model provenance, AIGovTool for hardware-enforced AI governance and usage controls, and BlindChat as a user-facing confidential conversational AI interface.

Product overview

Mithril Security offers a modular portfolio of open-source tools for confidential and traceable AI. The core platform is BlindLlama, a confidential AI inference framework that serves LLMs inside hardware-based secure enclaves with TPM-rooted cryptographic attestation, ensuring user data is never exposed even to the AI provider's administrators. Built on BlindLlama, the company offers three integrated modules: AICert (add-on) provides cryptographic model provenance and AI Bill of Materials to verify training data and code provenance; AIGovTool (add-on) enforces AI governance policies including usage controls and off-switch triggers for models deployed on untrusted infrastructure; and BlindChat (sub-brand) delivers a private conversational AI interface for end users, operating either locally in-browser or via the BlindLlama enclave backend. Together, the portfolio addresses the core privacy, traceability, and governance challenges of enterprise LLM adoption.

Differentiator

Problem solved

Functional benefit

Brands

• BlindLlama: Open-source framework for confidential and verifiable AI APIs with GPU support, using secure hardware and TPMs for data confidentiality
• AICert
• AIGovTool
• BlindChat
• BlindAI
• BastionLab

Products and services

• BlindLlama Open-source framework for deploying AI models on GPUs inside hardware-based secure enclaves with Trusted Platform Modules (TPMs) as the root of trust. AI models are served in hardened environments where data is never exposed even to the AI provider's administrators, and full code integrity is verified cryptographically. Functions as the confidential AI inference backend for BlindChat and enterprise deployments.
• AICert Open-source AI provenance solution that creates unforgeable model cards cryptographically binding a model hash to its training procedure hash using TPMs. Provides irrefutable proof of model origin, addressing copyright, supply-chain, and model-poisoning concerns.
• AIGovTool Open-source framework for AI model governance that enforces usage rules, consumption control, rate limiting, and off-switch triggers on AI models deployed on client infrastructure. Uses secure enclaves to protect model weights and provides trustable telemetry for AI custodians leasing models to untrusted parties.
• BlindChat Open-source private conversational AI application providing a user-facing interface to confidential AI. Runs in two modes: Local (small model entirely in-browser for full data locality) and Enclave (queries routed to BlindLlama backend with end-to-end encryption). Targets privacy-sensitive end users such as lawyers, accountants, and healthcare professionals.

Named customers5 records

Segments4 records

Ideal customer profiles4 records

Technology focussed Yes

API detail

Has API

No

API docs

API detail

Core technology

AI maturity

App detail

AI capability4 records

Feature5 records

Partnerships

Two partnerships are on record, tiered flagship and core.

• Future of Life InstituteflagshipStrategic or Co-development Partner · 1 November 2023Technical collaboration to develop AIGovTool, hardware-backed AI governance tools using Intel SGX enclaves. The partnership aims to enforce AI governance policies with cryptographic guarantees for model weight protection and controlled consumption.
• QuarkslabcoreOthers · 1 May 2023Independent security audit firm that conducted an in-depth security review of BlindAI and found no vulnerabilities, validating the security of Mithril's confidential computing approach.

Scale indicators2 records

Recent moves6 records

Expansion highlights6 records

Direct peers

• Anjuna Security: Anjuna provides a confidential computing platform that runs workloads inside hardware-isolated enclaves across major clouds. It directly competes with Mithril in confidential AI/ML workloads using secure enclaves and is the most head-to-head competitor in confidential AI infrastructure.
• Fortanix: Fortanix offers confidential computing and key management built on Intel SGX, with Confidential AI/ML capabilities for regulated workloads. It is comparable to Mithril as an enclave-based platform serving enterprise AI and data protection use cases.
• Edgeless Systems: Edgeless Systems builds open-source confidential computing platforms (Constellation, Contrast) using TEEs for AI and data workloads. It is a direct peer as an open-source confidential AI infrastructure vendor with comparable technology stack and target use cases.
• Opaque Systems: Opaque Systems provides a confidential AI/analytics platform built on secure enclaves and hardware-rooted attestation for collaborative data and ML. It competes with Mithril in the confidential AI/ML workload layer with similar enclave-based privacy guarantees.

Emerging players

• Phala Network: Phala offers confidential AI compute via TEEs on a decentralized network (Phat Contract, dstack). It overlaps with Mithril on confidential AI inference using secure enclaves and targets similar AI privacy use cases, with a decentralized go-to-market angle.
• Secret AI: Secret AI focuses on confidential LLM inference using secure enclaves and encrypted compute, targeting privacy-sensitive enterprise AI. It overlaps with Mithril's BlindLlama in confidential LLM deployment, with similar GPU-targeted positioning.

Broad incumbents

• Nvidia (Confidential Computing / Hopper Trusted Execution): Nvidia provides confidential computing capabilities on H100/H200 GPUs (e.g., TEE support) that directly intersect with Mithril's GPU-compatible confidential inference positioning. As the dominant AI hardware vendor, Nvidia's confidential GPU offerings could either compete or partner with Mithril's BlindLlama.
• Microsoft Azure Confidential Computing: Azure offers confidential VMs, Intel SGX/AMD SEV-SNP enclaves, and Confidential AI services for regulated workloads. It competes broadly with Mithril as the leading cloud-native confidential computing stack and bundles confidential AI with a much larger enterprise ecosystem.
• AWS Nitro Enclaves: AWS Nitro Enclaves provides isolated compute environments for processing sensitive data, including AI/ML workloads. Comparable to Mithril as a hyperscaler offering of enclave-based confidential compute, with significantly broader platform integration.

Others

• Hugging Face: Hugging Face hosts BlindChat and is the distribution channel for Mithril's conversational AI front end. While not a confidential computing peer, Hugging Face is adjacent ecosystem infrastructure for model distribution and the open-source AI community Mithril depends on.

Market position

Strengths5 records

Weaknesses5 records

Competitive moat4 records

Key risks6 records

Key highlights7 records

Customer concentration

• Xx.com/mithrilsecurity
• GitHubgithub.com/mithril-security
• YouTubeyoutube.com/channel/UCHUGBF1OSrTOil0UmeZ359w
• LinkedInlinkedin.com/company/mithril-security-company

Compliance1 record

Revenue estimate

Valuation estimate

Number of profiles

Profiles4 records

Funding overview

Funding rounds2 records

Investors4 records

Funding detail is available on the Subscription and Enterprise plan.Contact sales →

M&A

Investments

M&A and investment is available on the Subscription and Enterprise plan.Contact sales →

Frequently asked questions about Mithril Security