ITP Techblog

The US military's advanced research department (DARPA) wants to develop new encryption technology to protect privacy as masses of personal data move to cloud computing platforms.

DARPA's Data Protection in Virtual Environments (DPRIVE) programme is developing fully homomorphic encryption (FHE). This is an encryption system that allows data to be accessed, analysed and used to extract results, even while it remains fully encrypted.

It is seen as an answer to the myriad problems associated with data sharing in sensitive industries such as healthcare, finance and in government where data sharing even between government agencies is fraught with privacy challenges.

But if the world of FHE, you could have sets of encrypted hospital data being accessed to yield answers that could improve front line health care without any of that data being decrypted. 

The problem is that accessing encrypted data in real-time to run statistical analysis is hugely intensive from a computing standpoint. That's why DARPA has drawn chip maker intel into the programme to literally come up with designs of silicon processor that are up to the job. Microsoft, with its Azure cloud and JEDI cloud computing platforms, will help DARPA understand how HFE could work in the cloud, where data is increasingly hosted.

No decryption

Most data centres and cloud platforms hold and transfer their clients' data in encrypted form. But decryption is currently required at some point in the process to make sense of the data. Data sets are regularly compromised, which is leading to beefed-up privacy laws around the world, including here in New Zealand, where the revamped Privacy Act went into effect in December.

In the case of government agencies, researchers could be able to combine multiple population data sets to build models, without ever directly accessing or seeing the underlying data. Because homomorphically encrypted data is never decrypted, it means that data could also be physically held by a wider group of organisations.

The barriers to overcome to make full HFE a reality are huge. Compute times are 10,000 to 1 million times slower using this mention than decrypting data sets to access them. So hardware and software to speed up the process will need to be developed. 

"A computation that would take a millisecond to complete on a standard laptop would takes weeks to compute on a conventional server running FHE today," says DARPA programme manager, Tom Rondeau.

The plan is to try and reduce that time from weeks to seconds or milliseconds. 

From weeks to seconds

"We currently estimate we are about a million times slower to compute in the FHE world then we are in the plaintext world. The goal of DPRIVE is to bring FHE down to the computational speeds we see in plaintext. If we are able to achieve this goal while positioning the technology to scale, DPRIVE will have a significant impact on our ability to protect and preserve data and user privacy," Rondeau added.

The project with Intel and Microsoft is a multi-year collaboration and comes as DARPA devotes more attention to the rise of quantum computing, which it acknowledges is "raising questions about the durability of some of the most advanced data protection technologies".