# MLS minutes

## Protocol (Raphael/Richard)

Ekr: Aim for encrypting as much as possible. Revealing ContentType 
complicates analysis, and also lets malicious delivery service drop 
specific ContentType messages on the floor.
Raphael: Handshake messages are ordered, application messages are not.
Ekr: Dropping ContentType might require trial decryption.
DKG: Any metadata reveal leaks information about structure to delivery 
service subject to traffic analysis. Need to make this intention clear 
in architecture document. Not OK with revealing ContentType.
Ben: Not chartered to encrypt as much as possible, unlike TLS.
Karthik: We want keys for handshake to be different from keys from 
application data. Should encrypt sender and generation. See no strong 
reason to encrypt ContentType. Data messages do have some ordering.
Benjamin: Could do trial decryption and hide both. Could also encrypt 
ContentType under secret known to the group.
DKG: Be wary of PII.
RLB: Uncertainty about ContentType encryption.
Joel: Might be possible to infer messages based on timing. Traffic 
analysis will leak a lot.
DKG: We should still encrypt regardless of traffic analysis. Caution 
against "too bad, so sad, traffic analysis wins."
Benjamin: Should be out of scope for now. Let's keep ContentType in the 
clear.
DKG: If we keep ContentType in the clear we prohibit people who have the 
capability of mitigating against traffic analysis.
Emad: Handshake messages are add/remove/update? Updates are not ordered.
Joel: Make this an extension for the traffic analysis mitigations?
Ekr: Table for now.
(RLB will file an issue to prevent handshake message suppression.)

Joel: Why is old epoch not valid?
Raphael: (missed)
Karthik: Do not understand security implications. Need to understand 
invariants that are true for the tree at any state in time.
RLB: Consider simplifying algebraically. Not sure about current 
complexity implications.

## Federation (Emad/Raphael)

Ekr: Interested.
DKG: Interested -- in scope?
Sean: Yes, in scope at the crypto layer. Not at the application layer.
Jon: Could be complicated.
Colin: Interested.
DKG: More interested in different clients and different delivery 
services. Not saying it should not happen in the browser. If we focus on 
browser security, multi-party security might make security posture 
worse.
Karthik/RLB: (missed)
DKG: Not all devices are currently exposed to the delivery service -- do 
we want to change that?
Raphael: Working on documenting multi-device modes. If we're comfortable 
with mode where individual devices are hidden behind virtual devices, 
that can be incorporated into the federation model.
Karthik: Notion of members is an abstract concept in the model. Hiding 
devices is a great idea. If hiding devices at the federation level 
(Google or Facebook) then the story changes.

## Deniability (Sofia)

Raphael: There are some cheap ways we can build deniability into MLS.
Nalini: Deniability should be optional, e.g., for enterprise use cases.
DKG: Regulators who want this may be understaffed.
RLB: Needs to be optional.
Sean: Worth exploring.

## Security Analysis (Karthik)

MT: More rigorous process can slow things down (ref. QUIC).
RLB: Also felt pain in interop. Interested in more stability.

## Next Interim Planning

Will take place around Eurocrypt.