Quote from one of the authors of the research:
"The right way to think of what secure obfuscation allows is to create, *under many technical conditions*, software that has secrets built into it. These secrets are used by the software to compute output, and yet the secrets remain hidden even if an attacker obtains the entire machine-level code of the software, which of course the attacker could run and analyze. Thus the attacker would be able to *use* the secrets only in the way that the software allows, but not recover these secrets in any way beyond that. "
Perhaps you could create distributed autonomous organizations where there are nobody who has the keys to control it. Keys could perhaps be automatically generated and stored in the software itself, but nobody could access its funds or shut it down.