Gentry has developed a way to periodically clean the data to enable such a system to self-correct and be fully homomorphic. However, using it requires the system to be capable of a certain number of operations, currently beyond Smart's implementation. Gentry and his IBM colleague Shai Halevi have been experimenting with their own variant of Smart's approach, he says, and should announce results of their improvements to it later in the summer.

At the moment, Smart is adjusting the system's parameters to find out what works best. "For example, generating the keys was very slow; now we can do that better," he says. "It's like tuning a racing car; you tweak the engine and discover the tires need adjusting."

Predicting when that tuning will result in a technique ready for practical use is still impossible, says Smart, "but it will now run, and for people to be actually playing with a completely new method within one year of it first being presented is incredibly fast for cryptography." By contrast, he points out, a technique known as elliptic curve cryptography that is now used to secure mobile devices like the BlackBerry was first presented in 1985 but not implemented practically until around five years later.

Eleanor Rieffel, a senior research scientist at FX Palo Alto Laboratory, a research center at Fuji Xerox, agrees. "It has progressed fast, but because it's such a new area nobody really knows what route to take," she says. "These early implementations will let people experiment and try out ideas."

Meanwhile, despite the uncertainty over the idea's future development, interest from the IT world in any progress will remain high, says Rieffel. "There's more and more interest in being able to store data offsite with another company, or at a different site within a company, so this has a lot of attractions." It may even be that more powerful, but still limited implementations find use for specific applications, she adds.