search

dennisjackson / trust-negotiation-comments

2 stars 0 forks source link

Misunderstanding of identifiers used in Trust Anchor IDs #16

Open dadrian opened 1 month ago

dadrian commented 1 month ago

In "Fingerprinting and Client Privacy"

Trust Anchors improves on this design slightly, by having the client only expose the specific trust anchor it wants to use in this particular interaction with a website. [...] If the same identifiers are used between different root programs, then the privacy impact would be mitigated, but so would the utility of the signal.

There appears to be a misunderstanding. The identifiers in Trust Anchor IDs identify CAs used by the server, not root stores in use by the client. The CA identifiers are a global namespace interpreted (or ignored) by clients. See Section 8, Privacy Considerations. The utility of the signal is strongest when the namespace is global, as it would be a consistent identifier for what CAs are in use to all past, present, and future clients. I believe this may have been addressed recently in the draft: https://github.com/davidben/tls-trust-expressions/pull/46/files.

Please make sure your analysis reflects the latest version of the draft, and the full content of the Privacy Considerations section.

dennisjackson commented 1 month ago

The comments reflect the draft submitted for consideration at IETF 120. If you're not aware, IETF drafts are frozen during meetings to allow for meaningful discussion.