Using privacy-enhancing technologies to protect sensitive data
Privacy-enhancing technologies (PETs) are technologies that aim to protect privacy and confidentiality of data in use without reducing necessary system functionality. At Leidos, we are continuing to pursue the use of PETs as a means to address the White House Executive Order to Protect Americans’ Sensitive Data.
Last year we addressed several choice privacy technologies as areas for further research and outlined some of the major upcoming problems within the privacy space and how the government could best evaluate the disparate capabilities of existing PETs.
read more on Privacy-enhancing Technologies: Protecting Data in Use
This year, in partnership with AWS, we have explored security patterns for large language model (LLM)-enabled applications, specifically those that use personally identifiable information (PII) and protected health information (PHI). We selected AWS Nitro, a confidential computing suite, as our privacy mechanism to help mitigate the risks of sending PII/PHI to a third-party, LLM-enabled application.
Nitro Enclaves allowed us to deploy our application in isolation, safeguarding data in use from authorized access. This approach allows us to mitigate risks of exposing sensitive user data to a system provider during unencrypted LLM-based inference. This work dovetails nicely into our more general investigation into confidential computing and privacy for government services at large.
See our confidential computing approach
Leidos has also worked to develop repeatable, deployable solution patterns that help assure user privacy and system security for a broad range of applications. We have:
- Constructed a near real-time inferencing system for cloud-deployed, AI/ML models secured with fully homomorphic encryption.
- Developed a suite of tools for carrying out split learning at the edge, obfuscating model outputs while reducing datagram size.
- Stood up a Global Privacy Office to further our company’s policies and procedures with specific privacy and data measures.
We’re continuing to work diligently to produce new solution patterns, policies, and privacy assessments to address the needs of the American people and our government customers. Our growing body of work represents Leidos’ commitment to data privacy and the development of secure applications.
