Mithril Security provides a framework for AI model development that ensures data confidentiality and model integrity through cryptographic proof and secure hardware, such as Trusted Platform Modules (TPMs). This technology addresses the need for provenance traceability in AI, allowing users to verify that models are trained on trustworthy and unbiased datasets.
Funding
$1.3M raised to dateRaised to date based on public sources. This may differ from the amount the company actually raised and is based only on what is publicly available on the internet.
Founders
Product
Problem
AI models are vulnerable to various threats, including data poisoning, intellectual property theft, and compliance violations, due to the lack of transparency and traceability in their development and deployment. Verifying that AI models are trained on trustworthy and unbiased datasets remains a significant challenge.
Solution
Mithril Security provides a framework, AICert, that brings transparency and privacy to AI by enabling cryptographic proof of AI model provenance. AICert generates AI model ID cards that cryptographically bind model weights to the training data and code used to create them. By leveraging Trusted Platform Modules (TPMs), AICert creates non-forgeable proofs that attest to the entire stack used for producing the model, from the UEFI to the OS. This allows end-users to verify that the model they are interacting with originates from a specific training set and code, addressing copyright, security, and safety concerns.
Target Audience
Primary users are AI builders, including foundational model providers and companies that fine-tune models, as well as end-users seeking to verify the provenance and trustworthiness of AI models.
Features
- AI model traceability: Creates AI model ID cards that provide cryptographic proof, binding model weights to a specific training set and code.
- Non-forgeable proofs: Leverages TPMs to ensure non-forgeable AI model ID cards.
- Flexible training: Compatible with preferred training tools.
- No slowdown: Does not induce performance slowdown during training.
- Azure support: Integrates with Azure cloud services.
- AI Bill of Materials: Provides a detailed record of the data and code used in training.
- Network policy enforcement: Implements network policies to control data access and prevent unauthorized data exfiltration.