[Cryptography] Reading encrypted generative AI chats

[Kent Borg]

I only sort of follow your proposal for formalism, but I do like the 
idea that we might pull some of this messy external stuff into a well 
defined and tidy world where it could be cleaned up. But it will only 
get us so far, I think it is an uphill battle to pull more in.

I think cryptography is a victim of its own success. The "easy" (as in 
well defined) problems have been solved. Such as good symmetrical 
encryption, key exchange, public key encryption, quality hashing, etc. 
These are big wins, I remember times before all of those examples, when 
the public had no access to good cryptography and everything was in the 
clear or nearly so.

That big glaring problem has largely been solved, except cases when we 
just forget and don't even /try/ to secure data, sensitive and even 
boring data is encrypted pretty well these days. The next problems had a 
larger radius, such as the importance of encryption modes, and as the 
radius gets still larger to system issues, such as knowing what 
databases are in the system, whether RNG is even hooked up, how easily 
will users be tricked into giving over their access to con-men, etc., 
and cryptography has little to offer.

The above "we just forget" is the point.

The encryption algorithm isn't attacked anymore, not when it is so good, 
now that there are so many easier targets. Heck, even memory safety 
attacks are only ever used against a few fairly hard targets such as ios 
or Chrome, because there are so many easier ways, with ever more being 
installed constantly.

If your proposed formalism can create new well defined problems that are 
useful to solve, cryptography may live a bit longer. But it looks to me 
that cryptography is running low on problems to solve. Right?

-kb


P.S. On 3/19/24 04:55, Jerrold Leichter wrote:
> […] I note in particular Kent Boyd’s comment about […]

Correction: the name is Borg, as in "Björn" or "The" or a stupid way to 
drink too much too fast; not Boyd as in something different.



Jerry: This is edited from an earlier version you got but that I 
canceled before it got moderated.

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://www.metzdowd.com/pipermail/cryptography/attachments/20240319/8f615060/attachment.htm>