Mithril Security is a company working on confidential AI, which aims to keep AI models and the data they process protected while in use. Its website lists BlindLlama, which serves AI models on GPUs in a hardened environment so that the operator cannot see user data; AICert, which uses secure hardware to bind a model to its training procedure; and AIGovTool, a monitoring tool for AI systems. Its components are published on GitHub, and BlindLlama is licensed under Apache 2.0. The website does not state whether the services are vendor-hosted or deployed by the customer.
Mithril Security SAS is located at Ris-Orangis in France according to the legal notice on its website, and French law governs its website terms. Its privacy policy names a data protection officer but contains no statement on hosting regions or on GDPR compliance for its services. For a European organisation, a French provider with publicly available source code is relevant; the hosting location and contractual terms of any managed service should still be confirmed in writing.