Fujitsu cracks multi-source data encryption conundrum

Fujitsu Laboratories Ltd and Fujitsu Laboratories of America Inc. have announced the development of encryption technology that can match IDs or attribute values in information sources, such as classified or private data from multiple organisations, that are encrypted with different keys, without the need for decrypting the information.

As the use of the cloud and big data analysis has progressed, the demand for shared use of personal data and confidential information among multiple organisations has increased. In the healthcare sector for instance, there is a need to use clinical, health and genome information and tie it into the clinical studies or the drug-discovery business among multiple research organisations (figure 1).

Challenges in preserving

confidentiality

In such use cases, there are several methods for matching IDs and attributes while maintaining confidentiality. One is the hash function, which is a data transformation method widely used for checking whether passwords match, and another is homomorphic encryption, which enables addition, multiplication and searching of data while it is still encrypted.

With hash functions, it is difficult to restore original data but the same data is always transformed into the same value, so, when dealing with only a few data types, there is a possibility that the original data can be analogically inferred.

With homomorphic encryption, on the other hand, it is necessary for all organisations to use the same encryption key. While search results are encrypted, the key necessary to decrypt the search results can also decrypt all of the data, so it is necessary to strictly manage the key.

Fujitsu has developed the following techniques to work around these limitations.

Matching text strings encrypted with different

keys

Based on the theory of relational cryptography that allows calculation of the degree to which encrypted information matches, Fujitsu developed technology to determine a match between text strings encrypted with different encryption keys (figure 2). With this, registered strings and search strings are encrypted with the encryption key of each organisation. A registered string can be compared with the search string to see if they correspond while still encrypted, on a cloud server used for matching. The strings are encrypted with a one-way function, which is similar to a hash function, so they cannot be decrypted even with the keys used to encrypt them. The matching results are also encrypted and can only be seen by a person holding a dedicated match key.