I thought that polygraphs were not accepted as evidence in court, but are still used by some government and private agencies.

Also, just curious, have they been discredited as in that double-blind trials show no more than a 50/50 chance of being correct (i.e., flipping a coin)? Or is it that a high enough percentage of people will fail them even if innocent, or pass them even if guilty?

The thing I don't like about them is the method of reading them, where it is up to the subjective judgment of the examiner. That is, there isn't an algorithm that can be written down so that, for example, an untrained person or computer program can read it (and therefore be subjected to large enough random tests).