When you use factoring algorithms other than simple trial division - things like Pollard rho and its ilk - you are not guaranteed to get the smallest factor(s) first. Sometimes you get bigger factors, or combinations of factors, because those are what happen to pop out of the algebraic structure you're running over.

But any "primes" being used should have some serious factorisation algorithms run over them before being used, so this is really embarrassing.

EDIT

And to answer contravariant's question[0]:

    Also, shouldn't it only
    have 2 prime factors?

It's possible that one of the "primes" it found was in fact composite, and it's that "prime" that has these factors. So you generate two large primes and multiply them together, not realising that one of your "primes" wasn't.

[0] https://news.ycombinator.com/item?id=9561051

They used the GCD (greatest common divisor) algo to find any common divisors between 2 large keys. 231 was such a number.

This signals to me that both the keys are corrupted/bad to have really small prime factors (3, 7 and 11).

I have taken courses on discrete math, cryptography and read a few prime generation algos. Its a standard first step to check numbers against primes upto a million or so (nowadays, a billion). No prime generation algo I can imagine generated these keys.

Also, shouldn't it only have 2 prime factors?