*Weird quantum connections won't let you break Einstein's ultimate speed limit. But they will help you keep a secret*

The EPR paradox would be truly alarming - as opposed to just puzzling - if you could dabble with the spooky communication between the two particles to transmit an instantaneous message. If you could somehow choose the outcome of a measurement at your end, you would simultaneously engineer the outcome at the other end because of the way the two particles are entangled. The spooky connection would then let you send a message immediately to someone light years away. That would spell big problems for relativity theory, which strictly forbids any meaningful signals from travelling faster than the speed of light.

But Einstein can rest easy. This stratagem won't won't work because the result of your measurement can only ever be a fifty-fifty shot. You have no control over what is measured at your end, let alone the other.

That's not quite the end of the story, though. It's true you can't use the spooky connection to beat the speed of light, but you can use it to send ultra-secure coded messages. Not only that, but quantum theory actually lets you know if a spy has tried to tamper with your message.

Mathematicians have devised a variety of codes that depend on the use of a "key" - a number known only to the person sending a message and the recipient. Here's how it works. The sender transforms a written message into a set of binary digits, then scrambles the digits by applying a mathematical transformation using the key. The recipient receives the scrambled signal, undoes the mathematical transformation using the key, and thereby reconstructs the original message. Simple.

But there is a weak spot in this system. Both parties must decide on a key and then keep it absolutely secure. The key can be any random sequence of numbers, however, and that's where quantum theory can help. To exchange secret messages, Alice and Bob first set up a device that spits out entangled pairs of photons polarised the same way - either horizontally or vertically. One photon from each pair goes to Alice, and the other to Bob. To detect the photons, Alice and Bob each have a vertically polarised filter. If Alice sees a photon coming through her filter she knows that Bob has seen one too; if she doesn't see one, he won't either. Counting a detection as a "1" and a miss as a "0," both Alice and Bob record the same random string of binary digits which becomes their key to encode messages.

Now suppose a spy realises what Alice and Bob are up to, and inserts his own vertically polarised filter along the path to Alice's detector. The eavesdropper sneakily intercepts the information en route to Alice, and figures out the key. To cover his tracks, every time a photon passes through his filter he sends Alice an identical, vertically polarised photon which she naturally enough sees and records. If no photon goes through the spy's filter, he doesn't send one to Alice, so she can't pick one up. The upshot of this subterfuge is that the spy can intercept the photons to capture the key without either Alice or Bob knowing that their security has been breached.

With a slightly more complicated system, however, Alice and Bob can construct a key that's not only random, but is also immune to interception. If someone is poaching their photons, no matter how carefully, they would know about it. This time, our two protagonists opt for polarising filters that they can set either vertically or at 45 degrees. Every time a photon comes along, Alice and Bob set their filters randomly in either position. If, as will happen by chance half the time, they set their filters at the same angle, the situation is as before: when one sees a photon, so does the other.

But what if the filters are set at 45 degrees to each other? Suppose Alice's filter is vertical, and she sees a photon. Bob's photon must then be vertically polarised too. So what will happen when it encounters a filter set at a 45 degree angle? Classically, in such a case, the intensity of a beam of vertically polarised light would be cut in half. Translated into quantum terms, that means each photon in the beam must have a fifty-fifty chance of passing through. So Bob may or may not pick up the vertically polarised photon. It's a fifty-fifty guess.

Setting their filters at random each time, Alice and Bob measure a series of EPR photon pairs. Afterwards, they tell each other how their filters were set each time, by shouting across the rooftops if need be. If they both chose the same setting, they know they must have got the same result, and they can use the information to create a secret key. When their settings were different, neither can say what the other must have seen, so they ignore those results.

This may seem like a big step backwards. It has taken twice as many measurements to establish the key, since half of the measurements are thrown away. But here's the advantage: an eavesdropper cannot intercept the photons without giving himself away. A spy who wanted to insert his own detector to intercept Alice's photons would have to guess each time whether to set the filter vertically or at 45 degrees. If he guesses right he can pass on a counterfeit photon to Alice and no one will be any the wiser. But if he guesses wrong, he will send on the wrong type of photon to Alice. Alice will still detect photons and record results, but she and Bob can run statistical tests on their measurements that will betray the presence of a spy. This works because the photons Alice and Bob see are linked by the spooky EPR correlation, and eavesdropping destroys that link. In effect, the EPR correlation is now between Bob and the spy, and a quite different non-quantum correlation is introduced between the spy and Alice. The statistical tests tell Alice and Bob whether their EPR connection was preserved during the photon exchange, and hence whether their key is secure.

What all this means is that Alice and Bob can tell the world everything they are doing, except their own individual results for every measurement. They can set their cipher key without sending any sensitive information from one place to another, and they will know immediately if a spy has tried to crack their code.

A few years ago, Richard Hughes and colleagues at Los Alamos National Laboratory in New Mexico set up a system on a 14-kilometre optical fibre network that actually creates a secret cryptographic key. They haven't capitalised on their achievement for any nefarious purpose. At least, not as far as anyone can tell...