Skip to main content

ETH-MLS-OFFCHAIN

Abstract​

The following document specifies Ethereum authenticated scalable and decentralized secure group messaging application by integrating Message Layer Security (MLS) backend. Decentralization refers each user is a node in P2P network and each user has voice for any changes in group. This is achieved by integrating a consensus mechanism. Lastly, this RFC can also be referred to as de-MLS, decentralized MLS, to emphasize its deviation from the centralized trust assumptions of traditional MLS deployments.

Motivation​

Group messaging is a fundamental part of digital communication, yet most existing systems depend on centralized servers, which introduce risks around privacy, censorship, and unilateral control. In restrictive settings, servers can be blocked or surveilled; in more open environments, users still face opaque moderation policies, data collection, and exclusion from decision-making processes. To address this, we propose a decentralized, scalable peer-to-peer group messaging system where each participant runs a node, contributes to message propagation, and takes part in governance autonomously. Group membership changes are decided collectively through a lightweight partially synchronous, fault-tolerant consensus protocol without a centralized identity. This design enables truly democratic group communication and is well-suited for use cases like activist collectives, research collaborations, DAOs, support groups, and decentralized social platforms.

Format Specification​

The key words “MUST”, “MUST NOT”, “REQUIRED”, “SHALL”, “SHALL NOT”, “SHOULD”, “SHOULD NOT”, “RECOMMENDED”, “MAY”, and “OPTIONAL” in this document are to be interpreted as described in 2119.

Assumptions​

  • The nodes in the P2P network can discover other nodes or will connect to other nodes when subscribing to same topic in a gossipsub.
  • We MAY have non-reliable (silent) nodes.
  • We MUST have a consensus that is lightweight, scalable and finalized in a specific time.

Roles​

The three roles used in de-MLS is as follows:

  • node: Nodes are participants in the network that are not currently members of any secure group messaging session but remain available as potential candidates for group membership.
  • member: Members are special nodes in the secure group messaging who obtains current group key of secure group messaging. Each node is assigned a unique identity represented as a 20-byte value named member id.
  • steward: Stewards are special and transparent members in the secure group messaging who organize the changes by releasing commit messages upon the voted proposals. There are two special subsets of steward as epoch and backup steward, which are defined in the section de-MLS Objects.

MLS Background​

The de-MLS consists of MLS backend, so the MLS services and other MLS components are taken from the original MLS specification, with or without modifications.

MLS Services​

MLS is operated in two services authentication service (AS) and delivery service (DS). Authentication service enables group members to authenticate the credentials presented by other group members. The delivery service routes MLS messages among the nodes or members in the protocol in the correct order and manage the keyPackage of the users where the keyPackage is the objects that provide some public information about a user.

MLS Objects​

Following section presents the MLS objects and components that used in this RFC:

Epoch: Time intervals that changes the state that is defined by members, section 3.4 in MLS RFC 9420.

MLS proposal message: Members MUST receive the proposal message prior to the corresponding commit message that initiates a new epoch with key changes, in order to ensure the intended security properties, section 12.1 in MLS RFC 9420. Here, the add and remove proposals are used.

Application message: This message type used in arbitrary encrypted communication between group members. This is restricted by MLS RFC 9420 as if there is pending proposal, the application message should be cut. Note that: Since the MLS is based on servers, this delay between proposal and commit messages are very small.

Commit message: After members receive the proposals regarding group changes, the committer, who may be any member of the group, as specified in MLS RFC 9420, generates the necessary key material for the next epoch, including the appropriate welcome messages for new joiners and new entropy for removed members. In this RFC, the committers only MUST be stewards.

de-MLS Objects​

This section presents the de-MLS objects:

Voting proposal: Similar to MLS proposals, but processed only if approved through a voting process. They function as application messages in the MLS group, allowing the steward to collect them without halting the protocol. There are three types of voting proposal according to the type of consensus as in shown Consensus Types section, these are, commit proposal, steward election proposal and emergency criteria proposal.

Epoch steward: The steward assigned to commit in epoch E according to the steward list. Holds the primary responsibility for creating commit in that epoch.

Backup steward: The steward next in line after the epoch steward on the steward list in epoch E. Only becomes active if the epoch steward is malicious or fails, in which case it completes the commitment phase. If unused in epoch E, it automatically becomes the epoch steward in epoch E+1.

Steward list: It is an ordered list that contains the member ids of authorized stewards. Each steward in the list becomes main responsible for creating the commit message when its turn arrives, according to this order for each epoch. For example, suppose there are two stewards in the list steward A first and steward B last in the list. steward A is responsible for creating the commit message for first epoch. Similarly, steward B is for the last epoch. Since the epoch steward is the primary committer for an epoch, it holds the main responsibility for producing the commit. However, other stewards MAY also generate a commit within the same epoch to preserve liveness in case the epoch steward is inactive or slow. Duplicate commits are not re-applied and only the single valid commit for the epoch is accepted by the group, as in described in section filtering proposals against the multiple comitting.

Therefore, if a malicious steward occurred, the backup steward will be charged with committing. Lastly, the size of the list named as sn, which also shows the epoch interval for steward list determination.

Flow​

General flow is as follows:

  • A steward initializes a group just once, and then sends out Group Announcements (GA) periodically.
  • Meanwhile, each node creates and sends their credential includes keyPackage.
  • Each member creates voting proposals sends them to from MLS group during epoch E.
  • Meanwhile, the steward collects finalized voting proposals from MLS group and converts them into MLS proposals then sends them with corresponding commit messages
  • Evantually, with the commit messages, all members starts the next epoch E+1.

Creating Voting Proposal​

A member MAY initializes the voting with the proposal payload which is implemented using protocol buffers v3 as follows:


syntax = "proto3";

message Proposal {
string name = 10;                 // Proposal name
string payload = 11;              // Describes the what is voting fore 
int32 proposal_id = 12;           // Unique identifier of the proposal
bytes proposal_owner = 13;        // Public key of the creator
repeated Vote votes = 14;         // Vote list in the proposal
int32 expected_voters_count = 15; // Maximum number of distinct voters
int32 round = 16;                 // Number of Votes
int64 timestamp = 17;             // Creation time of proposal
int64 expiration_time = 18;       // Time interval that the proposal is active
bool liveness_criteria_yes = 19;  // Shows how managing the silent peers vote
}
message Vote {
int32 vote_id = 20;             // Unique identifier of the vote
bytes vote_owner = 21;          // Voter's public key
int64 timestamp = 22;           // Time when the vote was cast
bool vote = 23;                 // Vote bool value (true/false)
bytes parent_hash = 24;         // Hash of previous owner's Vote
bytes received_hash = 25;       // Hash of previous received Vote
bytes vote_hash = 26;           // Hash of all previously defined fields in Vote
bytes signature = 27;           // Signature of vote_hash
}

The voting proposal MAY include adding a node or removing a member. After the member creates the voting proposal, it is emitted to the network via the MLS Application message with a lightweight, epoch based voting such as hashgraphlike consensus. This consensus result MUST be finalized within the epoch as YES or NO.

If the voting result is YES, this points out the voting proposal will be converted into the MLS proposal by the steward and following commit message that starts the new epoch.

Creating welcome message​

When a MLS MLS proposal message is created by the steward, a commit message SHOULD follow, as in section 12.04 MLS RFC 9420 to the members. In order for the new member joining the group to synchronize with the current members who received the commit message, the steward sends a welcome message to the node as the new member, as in section 12.4.3.1. MLS RFC 9420.

Single steward​

To naive way to create a decentralized secure group messaging is having a single transparent steward who only applies the changes regarding the result of the voting.

This is mostly similar with the general flow and specified in voting proposal and welcome message creation sections.

  1. Each time a single steward initializes a group with group parameters with parameters as in section 8.1. Group Context in MLS RFC 9420.
  2. steward creates a group anouncement (GA) according to the previous step and broadcast it to the all network periodically. GA message is visible in network to all nodes.
  3. The each node who wants to be a member needs to obtain this anouncement and create credential includes keyPackage that is specified in MLS RFC 9420 section 10.
  4. The node send the KeyPackages in plaintext with its signature with current steward public key which anounced in welcome topic. This step is crucial for security, ensuring that malicious nodes/stewards cannot use others' KeyPackages. It also provides flexibility for liveness in multi-steward settings, allowing more than one steward to obtain KeyPackages to commit.
  5. The steward aggregates all KeyPackages utilizes them to provision group additions for new members, based on the outcome of the voting process.
  6. Any member start to create voting proposals for adding or removing users, and present them to the voting in the MLS group as an application message.

However, unlimited use of voting proposals within the group may be misused by malicious or overly active members. Therefore, an application-level constraint can be introduced to limit the number or frequency of proposals initiated by each member to prevent spam or abuse. 7. Meanwhile, the steward collects finalized voting proposals with in epoch E, that have received affirmative votes from members via application messages. Otherwise, the steward discards proposals that did not receive a majority of "YES" votes. Since voting proposals are transmitted as application messages, omitting them does not affect the protocol’s correctness or consistency. 8. The steward converts all approved voting proposals into corresponding MLS proposals and commit message, and transmits both in a single operation as in MLS RFC 9420 section 12.4, including welcome messages for the new members. Therefore, the commit message ends the previous epoch and create new ones. 9. The members applied the incoming commit message by checking the signatures and voting proposals and synchronized with the upcoming epoch.

Multi stewards​

Decentralization has already been achieved in the previous section. However, to improve availability and ensure censorship resistance, the single steward protocol is extended to a multi steward architecture. In this design, each epoch is coordinated by a designated steward, operating under the same protocol as the single steward model. Thus, the multi steward approach primarily defines how steward roles rotate across epochs while preserving the underlying structure and logic of the original protocol. Two variants of the multi steward design are introduced to address different system requirements.

Consensus Types​

Consensus is agnostic with its payload; therefore, it can be used for various purposes. Note that each message for the consensus of proposals is an application message in the MLS object section. It is used in three ways as follows:

  1. Commit proposal: It is the proposal instance that is specified in Creating Voting Proposal section with Proposal.payload MUST show the commit request from members. Any member MAY create this proposal in any epoch and epoch steward MUST collect and commit YES voted proposals. This is the only proposal type common to both single steward and multi steward designs.
  2. Steward election proposal: This is the process that finalizes the steward list, which sets and orders stewards responsible for creating commits over a predefined number of range in (sn_min,sn_max). The validity of the choosen steward list ends when the last steward in the list (the one at the final index) completes its commit. At that point, a new steward election proposal MUST be initiated again by any member during the corresponding epoch. The Proposal.payload field MUST represent the ordered identities of the proposed stewards. Each steward election proposal MUST be verified and finalized through the consensus process so that members can identify which steward will be responsible in each epoch and detect any unauthorized steward commits.
  3. Emergency criteria proposal: If there is a malicious member or steward, this event MUST be voted on to finalize it. If this returns YES, the next epoch MUST include the removal of the member or steward. In a specific case where a steward is removed from the group, causing the total number of stewards to fall below sn_min,
    it is required to repeat the steward election proposal. Proposal.payload MUST consist of the evidence of the dishonesty as described in the Steward violation list, and the identifier of the malicious member or steward. This proposal can be created by any member in any epoch.

The order of consensus proposal messages is important to achieving a consistent result. Therefore, messages MUST be prioritized by type in the following order, from highest to lowest priority:

  • Emergency criteria proposal

  • Steward election proposal

  • Commit proposal

This means that if a higher-priority consensus proposal is present in the network, lower-priority messages MUST be withheld from transmission until the higher-priority proposals have been finalized.

Steward list creation​

The steward list consists of steward nominees who will become actual stewards if the steward election proposal is finalized with YES, is arbitrarily chosen from member and OPTIONALLY adjusted depending on the needs of the implementation. The steward list size, defined by the minimum sn_min and maximum sn_max bounds, is determined at the time of group creation. The sn_min requirement is applied only when the total number of members exceeds sn_min; if the number of available members falls below this threshold, the list size automatically adjusts to include all existing members.

The actual size of the list MAY vary within this range as sn, with the minimum value being at least 1.

The index of the slots shows epoch info and value of index shows member ids. The next in line steward for the epoch E is named as epoch steward, which has index E. And the subsequent steward in the epoch E is named as the backup steward. For example, let's assume steward list is (S3, S2, S1) if in the previous epoch the roles were (backup steward: S2, epoch steward: S1), then in the next epoch they become (backup steward: S3, epoch steward: S2) by shifting.

If the epoch steward is honest, the backup steward does not involve the process in epoch, and the backup steward will be the epoch steward within the epoch E+1.

If the epoch steward is malicious, the backup steward is involved in the commitment phase in epoch E and the former steward becomes the backup steward in epoch E.

Liveness criteria:

Once the active steward list has completed its assigned epochs,

members MUST proceed to elect the next set of stewards (which MAY include some or all of the previous members). This election is conducted through a type 2 consensus procedure, steward election proposal.

A Steward election proposal is considered valid only if the resulting steward list is produced through a deterministic process that ensures an unbiased distribution of steward assignments, since allowing bias could enable a malicious participant to manipulate the list and retain control within a favored group for multiple epochs.

The list MUST consist of at least sn_min members, including retained previous stewards, sorted according to the ascending value of SHA256(epoch E || member id || group id), where epoch E is the epoch in which the election proposal is initiated, and group id for shuffling the list across the different groups. Any proposal with a list that does not adhere to this generation method MUST be rejected by all members.

We assume that there are no recurring entries in SHA256(epoch E || member id || group id), since the SHA256 outputs are unique when there is no repetition in the member id values, against the conflicts on sorting issues.

Multi steward with big consensuses​

In this model, all group modifications, such as adding or removing members, must be approved through consensus by all participants, including the steward assigned for epoch E. A configuration with multiple stewards operating under a shared consensus protocol offers increased decentralization and stronger protection against censorship. However, this benefit comes with reduced operational efficiency. The model is therefore best suited for small groups that value decentralization and censorship resistance more than performance.

To create a multi steward with a big consensus, the group is initialized with a single steward as specified as follows:

  1. The steward initialized the group with the config file. This config file MUST contain (sn_min,sn_max) as the steward list size range.
  2. The steward adds the members as a centralized way till the number of members reaches the sn_min. Then, members propose lists by voting proposal with size sn as a consensus among all members, as mentioned in the consensus section 2, according to the checks: the size of the proposed list sn is in the interval (sn_min,sn_max). Note that if the total number of members is below sn_min, then the steward list size MUST be equal to the total member count.
  3. After the voting proposal ends up with a steward list, and group changes are ready to be committed as specified in single steward section with a difference which is members also check the committed steward is epoch steward or backup steward, otherwise anyone can create emergency criteria proposal.
  4. If the epoch steward violates the changing process as mentioned in the section Steward violation list, one of the members MUST initialize the emergency criteria proposal to remove the malicious Steward. Then backup steward fulfills the epoch by committing again correctly.

A large consensus group provides better decentralization, but it requires significant coordination, which MAY not be suitable for groups with more than 1000 members.

Multi steward with small consensuses​

The small consensus model offers improved efficiency with a trade-off in decentralization. In this design, group changes require consensus only among the stewards, rather than all members. Regular members participate by periodically selecting the stewards by steward election proposal but do not take part in commit decision by commit proposal. This structure enables faster coordination since consensus is achieved within a smaller group of stewards. It is particularly suitable for large user groups, where involving every member in each decision would be impractical.

The flow is similar to the big consensus including the steward list finalization with all members consensus only the difference here, the commit messages requires commit proposal only among the stewards.

Filtering proposals against the multiple comitting​

Since stewards are allowed to produce a commit even when they are not the designated epoch steward, multiple commits may appear within the same epoch, often reflecting recurring versions of the same proposal. To ensure a consistent outcome, the valid commit for the epoch SHOULD be selected as the one derived from the longest proposal chain, ordered by the ascending value of each proposal as SHA256(proposal). All other cases, such as invalid commits or commits based on proposals that were not approved through voting, can be easily detected and discarded by the members.

Steward violation list​

A steward’s activity is called a violation if the action is one or more of the following:

  1. Broken commit: The steward releases a different commit message from the voted commit proposal. This activity is identified by the members since the MLS RFC 9420 provides the methods that members can use to identify the broken commit messages that are possible in a few situations, such as commit and proposal incompatibility. Specifically, the broken commit can arise as follows:
    1. The commit belongs to the earlier epoch.
    2. The commit message should equal the latest epoch
    3. The commit needs to be compatible with the previous epoch’s `MLS proposal`.
  2. Broken MLS proposal: The steward prepares a different MLS proposal for the corresponding voting proposal. This activity is identified by the members since both MLS proposal and voting proposal are visible and can be identified by checking the hash of Proposal.payload and MLSProposal.payload is the same as RFC9240 section 12.1. Proposals.
  3. Censorship and inactivity: The situation where there is a voting proposal that is visible for every member, and the Steward does not provide an MLS proposal and commit. This activity is again identified by the memberssince voting proposals are visible to every member in the group, therefore each member can verify that there is no MLS proposal corresponding to voting proposal.

Security Considerations​

In this section, the security considerations are shown as de-MLS assurance.

  1. Malicious Steward: A Malicious steward can act maliciously, as in the Steward violation list section. Therefore, de-MLS enforces that any steward only follows the protocol under the consensus order and commits without emergency criteria application.
  2. Malicious Member: A member is only marked as malicious when the member acts by releasing a commit message.
  3. Steward list election bias: Although SHA256 is used together with two global variables to shuffle stewards in a deterministic and verifiable manner, this approach only minimizes election bias; it does not completely eliminate it. This design choice is intentional, in order to preserve the efficiency advantages provided by the MLS mechanism.

Copyright and related rights waived via CC0

References​