[Cryptography] All applications need top security (was Re: Director GCHQ speaks at MIT)

Jerry Leichter leichter at lrw.com
Wed Mar 9 13:55:22 EST 2016


> It is seemingly reasonable to say that your discussion with a friend
> about what kind of beer to pick up for a party does not need the same
> level of protection as a dissident discussing an upcoming attempt to
> expose corruption in an election....
> 
> HOWEVER, the problem is that in practice, both activities will use
> exactly the same protocols and software, identically configured. You,
> as a protocol or software designer, do not get the luxury to provide
> "appropriate" levels of security for different uses. In practice,
> your protocols and software will sometimes be used for trivia and
> sometimes for things that incredibly important and you will have to
> design for the most important possible use....
This is a "yes, but" situation.  There may be some browser sessions you need to run through TOR, but running through TOR is inappropriate for anything that requires high throughput and/or low, predictable jitter.  There are large publicly available scientific data sets available through HTTP; moving them to HTTPS would require significant money to upgrade hardware.

In a world where you can always get the necessary money hence compute and other resources, there's little reason not to use the strongest crypto you can, uniformly.  But in a real, current, resource-constrained world, you sometimes need to make decisions.

Note that even the NSA, in previous iterations of Suite B, described different levels of security based on the material being protected (e.g., AES-128 at SECRET and below, AES-256 at TOP SECRET).  In the latest iteration, as hardware has gotten faster and cheaper, they pretty much dropped the distinction.  But the Suite B recommendations generally assume something with the power of, say, a modern laptop or better.  Dropping *all* such distinctions is another story.

*If* we have to maintain such distinctions, however, and important question is whether we can somehow give users a reasonable, understandable way to specify the level appropriate for a given connection/piece of data.  This is a *hard* job, and like most interesting UI questions related to security, has seen way too little useful work.
                                                        -- Jerry




More information about the cryptography mailing list