First & simplest: DOJ/FBI is made up of people, not of automatons. When you piss people off, you alter their behavior. Levison found himself in a situation where the welfare of his users depended in large part on how DOJ opted to behave. In order to make the most powerful statement he could about his dislike of DOJ's investigation, he deliberately provoked DOJ, spending his users welfare for his own PR ends.
Secondly, and I think equally importantly: Levison's actions made it easy for DOJ to demonstrate that he was untrustworthy. Their original demand was for metadata about a specific account, and the court record indicates that they were willing to spend time finding ways to cooperate with Levison, whose responses appear to have been inconsistent with his previous actions w/r/t/ court orders for Lavabit. By the end, no show of cooperation from Levison was needed, because Levison had made it clear to the court that DOJ had grounds to demand enough information for them to control the collection of the data they were after.
Yes, people with great responsibility, that should understand that someone may want to fight, deny or defend against legal actions taken by the DOJ. The fact that they're "human" is a flaw, as their actions should not be swayed by swings in emotion. The burden should be placed on them to act professionally and legally despite counteractions taken by those charged/targeted.
"Levison's actions made it easy for DOJ to demonstrate that he was untrustworthy"
Is Levison's character on trial here? Who give's a crap if he's untrustworthy, the only thing that should matter is the law. Sure in the court of public approval his character matters, but as with the first point it should not play a part in legal proceedings.
Ladies and gentlemen, behold! Eight words -- eight syllables -- laying bare why so many engineers tend to have such a hard time dealing with politics and people, on every scale from the picayune to the planetary. I've never seen it put so succinctly.
I'm still baffled as to why it's such a common problem, though. The essence of our shared discipline is to bridge the gap between a perfect idea and an imperfect world. This being the case, in our professional lives we must either be capable of recognizing and accepting such imperfection, or be incompetent; there is no third way.
So why is it that, in the interpersonal realm, so many of us prefer to demand impossible perfection than to exercise the same capacity that serves us so well in the technical work of our field? There's no sensible way of understanding how the imperfect human world works, save to begin by redressing ignorance through observation. Instead, many of us (some days, I think, most of us) flee into any number of idiosyncratic political fantasies -- just-so stories about how that world should work -- and refuse outright even to consider the possible value of discovering how that world does work.
Such refusal hampers us in any number of ways, professionally and otherwise; three of the former sort, right off the top of my head, are office politics, client interaction, and user interface design. Engineers are, of course, famously incompetent in all these areas, and unlike a lot of stereotypes, this one I've observed to be much more often accurate than otherwise. But I'm still struggling toward a conception of how this unhappy state of affairs comes to be.
When it comes to the DoJ and FBI, these institutions are immensely powerful compared to individuals, they should have MORE restrictions on an individual agent/employee taking retaliatory actions based on their emotional state.
The DOJ's emotional state has nothing to do with their ability to order Levison to hand over the keys because the order didn't come from them, it came from the Court.