Over recent years, the gap between theoretical security of quantum crytography and practical implementation has provided plenty of fun for super-geniuses the world over.
Yes, quantum cryptography is supposed to be unbreakable. After all, if anybody even observes the state of a qubit that Alice has prepared, entangled with another and sent to Bob, the entanglement is destroyed, and Bob will know something's wrong.
However, practical implementations of quantum cryptography left various possible attack vectors. To close these attacks (described in more detail below), the quantum crypto community proposed a new protocol, MDI-QKD (measurement device independent quantum key distribution), and now, two research groups working independently have verified that MDI-QKD gets a long way towards a provably-secure quantum crypto scheme.
One group worked out of Canada's University of Calgary (paper available at Arxiv, here), while the other was an international group comprising researchers from the University of Science and Technology, Hefei, Tsinghua University in Beijing, and Stanford University.
The scheme common to the two tests is to include a third party, Charlie, in the key-exchange process. First proposed by Hoi-Kwong Lo at the University of Toronto, the protocol asks Charlie to perform a single measurement on both Alice's and Bob's photons to determine whether their pulses are polarised at right angles to each other.
Importantly, the Charlie detector doesn't report on Alice's / Bob's polarisation – only the difference between their polarisations. Hence: if both Alice and Bob send vertically polarised pulses, Charlie will tell Bob “no”, Bob will adjust his polarisation, and Alice and Bob will use this as their key. Otherwise, Charlie will tell Bob “yes”, and the two ends will use their key without adjustment.
Since Charlie never reports polarisation values, all a third party (Eve) would be able to determine is whether Alice and Bob are synchronised. Eve can't tell from observing Charlie what the secret negotiated between Alice and Bob is.
The Canadian experiment took the MDI-QKD proposal on a field test – not using it to generate random keys, but to determine whether the measurement scheme would work over realistic distances. Charlie was kept on campus, while Alice and Bob were 6 km and 12 km away, respectively.
In the US-China test, Alice, Bob and Charlie were confined to the lab (albeit using a 50 km fibre on a reel): their test was demonstrating that MDI-QKD allows truly random keys to be generated. Not only that, but the test showed that realistic key generation rates of 25 kbit secure keys can be generated using the technique.
In both cases, the answer was “yes”. So while companies making commercial QKD kit had already started responding to the earlier attacks, there is now a protocol available for future designs. ®
Bootnote: Attack types
Let's look first at working with a single photon. If the eavesdropper, Eve, takes a guess at the polarisation Alice is sending, and gets it right, Bob will see a bright pulse from Eve and register it as a hit. If she gets it wrong, the avalanche photodiode at Bob's end would receive too dim a light to register anything at all – it would be a missed pulse and would count not as a “yes” or “no”, but as an error.
The problem here is that in older schemes, Bob might expect an error rate as high as 20 percent. That gives Eve enough opportunities to test her guesses before Bob decides the channel is considered to be compromised.
And no, El Reg is not aware of any successful real world attacks using these techniques.