Speaking of which (fat cats), it was nice to see a recent WSJ article ("Artificial Intelligence Isn't a Threat -- Yet" Gary Marcus) use "sandboxed" (influences limited in scope) in relation to the absence of "predict or control" within the context of computation (in a sense, we are already out of control). What we have, in essence, is test by trial (as in, luckless users trying to cope - even to the extent of automotive systems streaming out under very lax supervision?).
The issues raised in the article are central to truth engineering's core focus.
Now, about sandbox, it carries too toward those realms of high finance that have been so problematic (and, we will see another downturn - of worse consequences).
Remarks: Modified: 12/31/2014