One improvement I've seen is to place a medium value resistor (near 1k) from the PNP's emitter to its base. This helps to turn off the PNP and keep it off. This may partially compensate for NPN leakage since the small current could flow through that resistor rather then through the PNP's base.
Most transistors will leak to a small degree, even in reverse bias. For bipolar transistors leakages generally increase with higher temperatures. Such leakage is usually very small and is specified in the manufacturer's data.
In the case of your circuit, the NPN could have a small leakage from the collector to the base which could start to charge the capacitor. Adding a high value resistor across the capacitor may help discharge any such leakage to ground, however the resistor value should not be low enough that it significantly affects your intended turn on parameters.
To determine a maximum value for the added resistor (per the above) you would need to know the maximum leakage current from collector to base of the AC127, at the collector voltage being used. (from the Mfgr's spec's).
Then knowing the minimum base turn on voltage of the AC127, calculate the maximum value of the added resistor from: R = (V/I) - 100k, where "V" is the "Minimum Base Turn On Voltage, "I" is the CB leakage, the "100k" is included if the extra resistor is place across the capacitor. If the extra resistor were placed directly from the base to ground then don't include 100k in the calculation.
Having a resistor of that value or lower will prevent the leakage current from generating a base voltage that could turn on the transistor.
As to how or why it leaks, I believe it has to do with imperfections in the semiconductor material, thermal effects, and other quantum physic effects.