My two cents.
It seems that people defending Facebook's behaviour in this thread have collectively lost sight of what the point of a bug bounty is to begin with - to encourage people to report issues, rather than sell them.
We now have people arguing that "it is not acceptable to pivot beyond the initial intrusion for a bug bounty", even though a malicious attacker would have done the exact same thing. As long as standard no-damage rules are followed, where's the problem?
The bug bounty program is working exactly as intended, but the researcher is getting dinged over arbitrary rules. As somebody else here mentioned already: the reason blackhat work still pays, is because such arbitrary and bureaucratic rules do not exist there.
We should not forget that bug bounties are a tool, not a goal - the goal is to convince researchers to report rather than sell, and every part of a bug bounty and its rules must be designed accordingly.
Also: Why the hell were those AWS credentials not revoked immediately after compromise? This constitutes a grossly negligent failure on Facebook's part to assess impact, on top of their existing failure to have the "keys to the kingdom" on a single server to begin with.
And frankly, that failure only reinforces the need for the researcher pivoting into further systems, rather than just keeping it to a PoC - because evidently, nobody is going to assess impact at Facebook, if the researcher doesn't do it himself.