OK, so you're going with this scenario:

1. I know Apub and Bpub,
2. I know M is 3
3. I'm choosing a random number for C's private key

Cpub is g^C

The equation I am solving for .. and trying to factor myself out of is g^Ax + g^B*2 + g^C*3

I don't know A or B... I only know their public keys.

I don't think it's possible to adaptively choose C for an attack on the multisig construction, when using hash of the public key as the X coordinate in the polynomial, because in order to satisfy the equation and factor out C, you would need to be able to break the hash.

With an additive construction, yes... adaptive attacks are possible.   But in a shamir secret sharing interpolation, you need a public X coordinate as well as a secret share.   Choosing hash(pub) as X, prevents this attack.


On Wed, Jul 11, 2018 at 6:35 AM, Adam Back <adam.back@gmail.com> wrote:
On Wed, Jul 11, 2018, 02:42 Erik Aronesty via bitcoin-dev <bitcoin-dev@lists.linuxfoundation.org> wrote:
> Basically you're just replacing addition with interpolation everywhere in the musig construction 

Yes, but you can't do that without a delinearization mechanism to prevent adaptive public key choice being used to break the scheme using Wagner's attack. It is not specific to addition, it is a generalized birthday attack.

Look at the delinearization mechanism for an intuition, all public keys are hashed along with per value hash, so that pre-commits and forces the public keys to be non-adaptively chosen. 

Adaptively chosen public keys are dangerous and simple to exploit for example pub keys A+B, add party C' he chooses C=C'-A-B, now we can sign for A+B+C using adaptively chose public key C.

Btw Wagner also breaks this earlier delinearization scheme S=H(A)*A+H(B)*B+H(C)*C

Adam