IBM Discovers Scheme To Manipulate Encrypted Data

A Researcher at IBM has found an encryption method that lets you operate on data without decrypting it: good news for the cloud and spam filters

A researcher at IBM claims to have solved a major IT security conundrum, by constructing an encryption system which can manipulate encrypted data without encrypting it. The scheme could help secure cloud computing and fight encrypted spam.

The challenge of manipulating data without exposing it has bugged cryptographers for decades. But in a breakthrough, IBM researcher and Stanford University PhD candidate Craig Gentry has developed a “fully homomorphic encryption” scheme that keeps data protected.

“Basically, an annoying property of encryption is that it typically traps data inside a box that prevents the data from being used or analyzed until you open the box with the secret decryption key,” Gentry said. “What a fully homomorphic encryption scheme allows you to do is analyze or compute a function of the data while it remains securely inside the box.

“For example, suppose you want to store your files on an untrusted server,” he continued. “You don’t want the server to see your files; so, you want to encrypt them. On the other hand, you want to be able to access your files in an ‘intelligent’ way … There is a way to express this query as a function f. If your files are encrypted with the fully homomorphic encryption scheme, you can send your query to the server, which expresses it as some function f; the server then homomorphically computes an encryption of f(m_1, …, m_t) — where f(m_1, …, m_t) outputs the relevant files—and sends this ciphertext back to you. You decrypt to recover the files.”

It is also possible to encrypt the query as well, he said.
According to IBM, Gentry’s solution could help strengthen the business model of cloud computing when vendors are hosting confidential customer data, by enabling them to perform computations on data at their clients’ request without exposing the original data.

It also represents a potential boost for spam filtering of encrypted e-mails. Spam filtering, he noted, doesn’t need to be done online because of the relative asynchronicity of e-mail, and because the amount of data is likely to be small.

“The idea for spam-filtering encrypted e-mails is as follows: I encrypt e-mails to you under your public key pk using the fully homomorphic encryption scheme. Maybe some spammers also send you some e-mails encrypted under pk. Now, a spam filter can be expressed as some function f that is applied to (unencrypted) e-mail data, which outputs a ‘0’ if it is spam, and a ‘1’ if it is not,” Gentry explained.

“To spam-filter encrypted e-mails, the spam filter uses the fully homomorphic encryption scheme to compute a ciphertext that encrypts f(m), where the encrypted e-mail is the encryption of m. The filter prepends this ciphertext to the encryption of m. Now, when you check your encrypted e-mail, you first decrypt the prepended ciphertext. If the ciphertext encrypted a ‘0’, indicating the message is spam, you don’t bother decrypting the rest of the e-mail. Otherwise, you decrypt the rest.”

At this point, Gentry said, the next step is to work on improving the efficiency of his discovery. He admitted it takes many times longer to process data homomorphically while it is in the encryption box than it would to process the data in the clear.

“Before it becomes a tool, more theoretical work will need to be done to make it more efficient,” he said. “But now that researchers know that fully homomorphic encryption is possible and have an actual construction that they can sink their teeth into, I think there is reason to be optimistic that the efficiency will be improved dramatically.”

The Wikipedia page on homomorphic encryption explains that Gentry’s scheme usees lattice based cryptography, and can in principle perform any number of additions and multiplications on a set of data – with the one proviso that the maximum number must be set in advance when the key is generated, and the public key grows linearly with the number of multiplications allowed.