[TLS] Re: PKI dynamics and trust anchor negotiation

[Dennis Jackson <[e-mail]>]

I don't think there's any new argument to address, so I will just offer 
a pointers into where these issues are discussed in 'Trust is 
non-negotiable'.

Section 3.3 [1] looks at how trust negotiation (as an abstract 
mechanism) changes incentives to divergence for existing root programs, 
as well as the consequences of enabling the deployment of new root 
programs with divergent policies. Unfortunately, fragmentation is often 
an insidious, gradual process that arises from many actors making 
independent decisions that prioritize their own needs, in the absence of 
any ecosystem forces strong enough to bring them together. It doesn't 
have to be a goal that is pursued deliberately, although it can also be 
(e.g. market differentiation or enshittification).

Separately, structural barriers to deployment of a new technology, which 
impact certain constituencies much more than others, are a common cause 
of ecosystem fragmentation. The specific design of TAI has serious 
issues in this regard which are laid out in section 4.6 [2].

The argument that this fragmentation is possible with existing 
mechanisms like certificate_authorities is evaluated at the end of the 
section 3.3 [3]. These extensions do not have any meaningful 
existence in the wild for the purposes of server certificate negotiation 
and it's running code that counts here. Further, for the same reasons 
that the TAI authors identified these extensions as an unsatisfactory 
solution for their needs, it is unsuitable for anyone else trying to 
deploy trust negotiation at scale.

Finally, as is discussed throughout Section 3, I disagree there is any 
security tradeoff at the heart of this issue. We've shipped many many 
improvements over the past 10+ years through the steady ratcheting of 
root program policies. Instead, I believe barriers to further 
improvements are predominantly found server-side, in the level of 
investment that server operators (and to a lesser extent CAs) are 
willing to make in terms of libraries, automation and tooling. Trust 
negotiation does nothing to address these issues, which are largely 
societal rather than technical.

Best,
Dennis

[1] 
https://datatracker.ietf.org/doc/html/draft-jackson-tls-trust-is-nonnegotiable#section-3.3

[2] 
https://datatracker.ietf.org/doc/html/draft-jackson-tls-trust-is-nonnegotiable#section-4.6

[3] 
https://datatracker.ietf.org/doc/html/draft-jackson-tls-trust-is-nonnegotiable#name-alternative-paths-to-abuse