[Cryptography] Aside on random numbers (was Re: Opening Discussion: Speculation on "BULLRUN")

[Jerry Leichter]

On Sep 6, 2013, at 10:03 AM, Perry E. Metzger wrote:
> 
> Naively, one could take a picture of the dice and OCR it. However,
> one doesn't actually need to OCR the dice -- simply hashing the
> pixels from the image will have at least as much entropy if the
> position of the dice is recognizable from the image....
> 
> One could write an  app to do this, but of course the phone is
> not exactly a secure platform to begin with...
Ah, but that highlights an essential difference between OCR'ing the image and just hashing it:  I can easily check, with my own eyes, that the OCR app is really doing what it claims to be doing.  I have no hope of checking the hash-based app.  A whole class of attacks is closed off by the OCR technique.

It's not that there aren't other attacks.  The phone could, for example, leak the generated values, sending them off to Big Brother.  That kind of attack would, if done correctly, be virtually impossible to detect.  On the other hand, it's not nearly as valuable as a biased generation attack - Big Brother would receive streams of random die tosses with little context about what the resulting values would be used for or how they would be used.  Appropriately targeted attacks might work - "I know Metzger regenerates his keys on the 3rd of every month at about 8:00 AM, so let's use the values he scans at around that time as guesses for his base random values" - but we're talking quite a bit of difficulty here - and the more people use the app, and the more often they make it a habit to toss and scan dice and just discard the results, the more difficult it becomes.
                                                        -- Jerry