You don't see a difference in protection posture between something which has to be used in the wild vs something which could be kept in a vault for it's entire lifetime? I'm not arguing that escrow is a good idea, I'm arguing that this argument is not why that's true.

Edit: consider how often keymat has leaked.

There isn't one really - if they had access to it the same logistical issues would come up no matter how much they pinky promise not to abuse it. The more it spreads the more vulnerable it is

The only way key escrow would be remotely trustworthy would be if there were "hostages" to provide an intrinsic punishment for their failures or stonewalling of transparency.

You really think the keys will stay in a vault?

And you think this because you have assurances from who? The NSA? Xi Jinping? MBS? Kim Jung Un? Diane Feinstein? Donald Trump? RSA Inc.? NIST? David Sternlight?

I mean, there is a big difference - you have to directly send your exploits to the enemy so that they'd work; you don't have to send any escrowed keys to them.

The enemy will have escrowed keys too.

But see my answer to Gödel_unicode.

Fun fact, the need for NSA exploits to be installed on an “enemy” (target) device had nothing to do with how they were leaked in this case.

And stuff kept in vaults can be leaked too.

Sure the two things are different but the dynamics are the same. Information wants to be free, and it leaks.

Your argument can be turned around as well. The NSA tools were meant for use by a small specialized team under strict controls. Whereas key escrow has been put forward as a solution that would allow any podunk sheriff in any jurisdiction to apply for access to decrypt private information anywhere... virtually guaranteeing compromise with poor security practices, unlike what would be expected for the NSA tools.