SBIR Phase I: The Encore Oblivious Computation Framework

The broader impact of this Small Business Innovation Research (SBIR) Phase I project is to provide a fast, universal, and easy-to-program technology called Encore, that enables privacy-preserving computation on sensitive data. Users are becoming increasingly aware of the privacy risks of giving up private data to online services and sending sensitive queries to Large Language Models (LLMs). Further, there is a recent push by big players in industry (e.g., Google, Apple, Signal) to roll out new privacy-preserving services. The proposed Encore framework will allow businesses to “switch on” privacy for their existing services with minimum migration cost and runtime overhead. Further, it will help companies with existing privacy offerings to scale up their privacy-preserving services to big data and save computational cost. Through the company’s open-source efforts, the proposed project will make confidential computing techniques accessible to even non-expert programmers. This will in turn encourage wider adoption of confidential computing techniques, and pave the way for a private data economy where users are in full control of their data and may choose to contribute them to data analytics tasks (e.g., clinical or population-wide studies). This Small Business Innovation Research (SBIR) Phase I project will develop oblivious computation techniques that allow provably secure obfuscation of access patterns to sensitive data, while minimizing the overhead. Encryption at rest is a standard technique for protecting confidential data. However, encryption alone fails to hide the access patterns to data, which can completely reveal the users’ queries or intentions in many applications such as contact discovery, database search, and queries to Large Language Models. Oblivious computation relies on algorithmic techniques to “randomize” the access patterns such that they leak nothing. Earlier, the team proposed simple and practical oblivious computation techniques which have already gained large-scale adoption, e.g., by the encrypted messenger Signal. The proposed project will build on the team’s prior expertise and develop a new family of oblivious algorithms specifically optimized for hardware enclaves. Further, this project will develop new algorithmic techniques for parallelizing oblivious computation, as well as compilation techniques for converting insecure legacy code into oblivious implementations. The team plans to open source an Oblivious STL library which contains oblivious counterparts of common data structures and utility algorithms, and can be viewed as a privacy-preserving counterpart of the standard STL library for popular languages. This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.