It's funny how many different ideas are perfectly plausible in this scenario.
Basically, you've got very strong positive feedback and since the base current of Q1 is determined by the col. current of Q2, overall amplification is multiplication of both Betas of each transistor (could be well over 9000
).
Keep in mind that even 1nA at the Q2 base will make about 0,2uA at the Q1 base and that's gonna switch on whole circuit.
It doesn't really matter if it's load, the C-E capacity of Q1 (or Q2), Q2 base picking up some rf (or even mains) signal, etc.
The solution is the same, to
break the high gain loop. Probably best way with resistor between B and E of the Q1, something like 10k - 47k should do it. It's gonna limit the minimum current going through Q2 that is able to switch on whole circuit (about 0,5V/47k - B-E voltage drop over the BE resistor).