Scalable Certification of Confidential Assets

Starting a collaboration between entities usually requires checking whether the assets obtained from collaborators satisfy an entity’s expectations and use case requirements. For example, financial technology startups developing new fraud detection techniques need to ensure that the banks they collaborate with supply transaction data with certain conditions in a certain format. Similarly, pharmaceutical companies developing new drugs need to ensure that the hospitals they collaborate with supply patient data satisfying certain conditions. 

In certain industries, checking compliance requirements of regulatory and legal frameworks involves similar actions. These checks usually cover various properties of the assets in question. Examples of such properties and transformations include 

• ML datasets having appropriate statistical properties, relevant features and satisfactory measures taken for privacy, 
• ML models possessing satisfactory performance and matching lineage for used training datasets, 
• Build environments for software binaries whose source-code may not always be available, or reproduction is resource-heavy and time-consuming.

The challenges of trust between entities when exchanging information about the properties of their assets today are usually addressed through contractual agreements among collaboration partners and manual audits by trusted third parties. Creating these contracts and performing these audits are manual, costly, and time-consuming. Efforts in improving automation and security while sharing software and ML models, such as Software Bill of Materials (SBoMs) and Machine Learning Bill of Materials (MLBoMs), usually focus on increasing accountability by creating a provenance trail of the assets being used. 

In our Scalable Certification of Confidential Assets solution, we address trust challenges by providing verifiable evidence about the properties and transformations of assets while protecting their confidentiality. This evidence is third-party verifiable, meaning that third party entities can check its correctness. The entities can include not only the recipients or users of the assets in a collaboration but also auditors and regulators. The verifiable evidence is also self-contained, so that no additional infrastructure is needed than what already exists today. As a result, our solution improves compliance with company policies as well as privacy and security requirements of regulatory and legal frameworks. 

Our Scalable Certification of Confidential Assets solution leverages Trusted Execution Environments (TEEs), readily available in major Cloud Service Providers. Our solution creates third-party verifiable evidence about the properties of different types of confidential assets (e.g., datasets, models, software) in a manner that is cost-effective, automated and secure.

As a result, we enable the creation of trustworthy collaboration ecosystems, improving compliance and security in marketplaces and software supply chains, without sacrificing confidentiality.

Our people