Mithril Security
Mithril Security, founded in 2021 and headquartered in Ris-Orangis, France, was acquired by H Company in June 2024 and now operates as a subsidiary.
Profile
Mithril Security provides open-source tools that use confidential computing hardware to keep data and AI model code encrypted while in use, and to cryptographically prove how a model was trained.
Mithril Security, founded in 2021 and headquartered in Ris-Orangis, France, was acquired by H Company in June 2024 and now operates as a subsidiary. The company raised approximately €1.2 million (about $1.3 million) in pre-seed and seed funding from Auriga Cyber Ventures, Polytechnique Ventures, ITFarm, and the UC Berkeley SkyDeck incubator. At the time of acquisition, it employed 13 people.
Mithril Security develops open-source tools for confidential AI deployment, primarily BlindLlama (formerly BlindBox) for deploying large language models with hardware-based data confidentiality, and AICert for cryptographically proving a model's training provenance using secure hardware like TPMs. Its products are designed to protect both user data and model intellectual property when AI is run in cloud or third-party environments. The company's technology has been audited by independent security lab QuarksLab.
Customers include the Future of Life Institute, a stealth healthcare AI startup working with the NHS, Avian digital forensics, Auxlia (airport security AI), and Scille (editor of PARSEC). Mithril Security was also a recipient of the OpenAI Cybersecurity Grant Program. Post-acquisition, the company has not disclosed standalone revenue or further funding rounds. Its market position is as a niche provider of confidential computing tooling for AI, now folded into H Company's broader portfolio.
Who buys this
- Healthcare organizations deploying AI on sensitive patient data while needing regulatory compliance
- Digital forensics and investigation firms analyzing confidential documents with LLMs
- Airport security and physical safety companies running AI models on-premise to protect intellectual property
- Enterprises building zero-trust data sharing solutions on public cloud infrastructure
- AI providers and enterprises that need to prove model provenance and training integrity to regulators or customers
Publicly disclosed clients
- Future of Life Institute
- Avian (digital forensics)
- Auxlia (airport security AI)
- Scille (editor of PARSEC)
- Stealth healthcare AI startup (working with NHS)
Strengths and what to watch
Strengths
- Open-source codebase (GitHub) and independent security audit by QuarksLab provide transparency and third-party verification.
- Hardware-backed confidential computing (TPMs, secure enclaves) offers cryptographic proof of data confidentiality and model provenance, a differentiator from pure software encryption.
- Recipient of OpenAI Cybersecurity Grant Program, lending credibility and visibility in the AI security ecosystem.
Watch for
- Acquired by H Company in June 2024; as a subsidiary, Mithril Security's product roadmap, brand independence, and ability to raise further capital are now controlled by the parent.
- Very small company (13 employees at acquisition) with limited disclosed revenue; customer concentration risk is high given only a handful of named clients.
- Confidential computing for AI is a niche within a niche; competition from larger platforms (e.g., Fortanix, Opaque) and cloud providers' native confidential computing offerings could limit market share.
Key Information
- Industry
- AI Observability
- Founded
- 2021
- Headquarters
- Ris-Orangis, France
Frequently Asked Questions
What is Mithril Security?
Mithril Security is a French company founded in 2021 that provides open-source tools for confidential AI deployment. It uses hardware-based confidential computing to keep data and AI models encrypted while in use and to cryptographically prove how a model was trained.
What products does Mithril Security offer?
Mithril Security offers BlindLlama, formerly BlindBox, for deploying large language models with hardware-based data confidentiality, and AICert for cryptographically proving a model's training provenance using secure hardware like TPMs. Both are open-source and available on GitHub.
Who are Mithril Security's typical customers?
Mithril Security serves healthcare organizations handling sensitive patient data, digital forensics firms analyzing confidential documents, airport security companies protecting AI intellectual property, and enterprises needing zero-trust data sharing on public cloud infrastructure.
How does Mithril Security ensure data confidentiality?
Mithril Security uses confidential computing hardware like TPMs and secure enclaves to encrypt data and AI model code while in use. This hardware-backed approach provides cryptographic proof of data confidentiality, differentiating it from pure software encryption methods.
Was Mithril Security acquired?
Yes, Mithril Security was acquired by H Company in June 2024 and now operates as a subsidiary. The company had raised about €1.2 million in pre-seed and seed funding from investors including Auriga Cyber Ventures and UC Berkeley SkyDeck.
What security audits has Mithril Security undergone?
Mithril Security's technology has been audited by independent security lab QuarksLab, providing third-party verification of its security claims. The company also received the OpenAI Cybersecurity Grant Program, adding credibility in the AI security ecosystem.
Sources
- www.cbinsights.com — Total raised $1.7M, acquisition by H Company on June 1, 2024, investor list
- pitchbook.com — Acquired/Merged status, 13 employees, parent company H Company, primary industry
- blog.mithrilsecurity.io — €1.2M pre-seed round led by Auriga Cyber Ventures, Polytechnique Ventures, ITFarm, UC Berkeley SkyDeck; product BlindBox/BlindLlama details
- www.mithrilsecurity.io — Product descriptions (AICert, BlindLlama), customer testimonials (Future of Life Institute, Avian, Auxlia, Scille, stealth healthcare startup), OpenAI Cybersecurity Grant, QuarksLab audit
- tracxn.com — Total raised $1.3M, seed round May 17, 2023, investor list