End-to-end encryption without verifying the authenticity of the exchanged public long-term keys only enables the endpoints to protect their communication against passive attacks. This means an attacker cannot read encrypted messages in transit without actively intervening in the key exchange. However, without any other precautions, active attacks are still possible. If an attacker replaces the exchanged keys with malicious ones or introduces an additional malicious endpoint, the end-to-end encrypted messages can be read and manipulated by the attacker.
When using end-to-end encryption where public long-term keys are transmitted over a channel which is not protected against active attacks, the authenticity of those keys is not guaranteed. Such a key has to be authenticated by the receiving endpoint over another channel which is already protected against active attacks to maintain the confidentiality of sent messages and ensure the authenticity and integrity of received messages. Trust messages can be used to transfer the needed data via XMPP for performing such an authentication. Furthermore, they can transmit the data used for distrusting a key.
Trust messages can be used in conjunction with an end-to-end encryption protocol such as OpenPGP for XMPP (XEP-0373)  or OMEMO Encryption (XEP-0384)  to automatically or semi-automatically establish secure channels protected against active attacks.
Trust messages have the following advantages:
A permanent storage is not needed. Cached data can be be removed as soon as the recipient made use of them. E.g., a server-side storage holding available data because it does not know when a client needs to access them, as for certificate-based approaches, is unnecessary.
An infrastructure in additional to the one used for messages is not needed because trust messages are ordinary messages. E.g., a server-side storage and a mechanism for accessing it other than those used for messages is unnecessary.
The authenticity and integrity of trust messages are ensured by a signing mechanism. If trust messages are additionally encrypted, they have the following advantages:
The cryptographic properties of the encryption protocol are applied to the trust messages. Properties such as confidentiality, forward secrecy and deniability can have a positive impact on the authentication of keys. Hence, trust messages, in contrast to certificates, can e.g. be deniable toward a third party if the encryption protocol provides that kind of deniability.
The fact that an endpoint trusts a key or not can be kept confidential toward an attacker by encrypting those messages and sending them only to endpoints with authenticated keys. This means that an attacker cannot detect by the content of a trust message whether a specific key is trusted by the sender. If the trust message is sent for an authentication of a key, the involved parties of that authentication will therefore stay anonymous toward an attacker.
The encryption protects against passive attacks on the transmission of the trust messages. That way, an attacker cannot read the content of the trust messages.
The restriction to send trust messages only to endpoints with authenticated keys in addition to the encryption protects against active attacks on the transmission of the trust messages. An attacker will even after introducing a malicious key not receive a trust message encrypted with that key.
Blocking data used to trust or distrust keys in transit is made more difficult. In particular, an attacker is not able to selectively block transmitted data used to distrust the attacker's key.
If an attacker cannot distinguish whether the data sent from a client is used for trusting or distrusting a key, the attacker can only randomly block some messages or the whole communication. If the communication is already compromised by an active attack, the attacker does not want to stop the whole communication. During that state, the attacker has the possibility to keep on e.g. eavesdropping or altering messages. Therefore, the attacker wants to block data which can lead to excluding the attacker. But the attacker does not want to block the communication itself.
Data which is used by the recipient to distrust the attacker's key would make it impossible for the attacker to continue to encroach on the communication. Thus, it is important to prevent an attacker from blocking data used for making trust decisions. E.g., an approach using certificates permanently stored on a server cannot prevent an attacker from specifically blocking such data because certificates have to be discoverable and identifiable as such.
It might be possible to distinguish an encrypted trust message from other encrypted messages by analyzing the network traffic over a period of time. However, the mitigation of that issue is out of scope.
A trust message (i.e., the root <message/> element)
The last two points are needed to achieve their mentioned goals because a trust message does not contain a <body> element which would automatically lead to the desired result.
In the following example, two OMEMO Encryption (XEP-0384)  keys of Alice are indicated as trusted, one key of Bob is indicated as trusted and two other ones of Bob are indicated as untrusted.
Trust messages MAY be sent unencrypted to solely communicate the trust in specific keys without any other protection. But as described before, the strength of trust messages is the possibility to encrypt their content and choose to which endpoints they are sent according to the trust in the endpoints' keys.
The <trust-message/> element MUST be signed before sending for ensuring the authenticity and integrity.
The <trust-message/> element SHOULD be signed and encrypted before sending for applying all advantages the encryption provides. That protects against passive attacks on the transmission of the trust message. The trust message SHOULD only be sent to endpoints whose keys have already been authenticated for also preventing active attacks on the transmission of the trust message.
Describing how the <trust-message/> element has to be used by each existing encryption protocol is out of scope. Stanza Content Encryption (XEP-0420)  specifies a common method for encrypting arbitrary elements which can be used by different encryption protocols. When using an encryption protocol such as OMEMO Encryption (XEP-0384)  which uses Stanza Content Encryption (XEP-0420)  (SCE), the SCE <payload/> element MUST contain the <trust-message/> element as a direct child.
A trust message SCE <content/> element
Message Carbons (XEP-0280)  minimizes the number of trust messages to be sent while having the same payload because trust messages with the same payload do not have to be sent for each endpoint. In combination with the usage of Message Archive Management (XEP-0313) , the delivery of trust messages to temporarily offline endpoints is ensured even if they are available under a different resource after going online than the last known one before going offline.
This specification uses Message Carbons (XEP-0280)  for sending a trust message to all endpoints of a contact or to all own endpoints at once. By sending a trust message to the contact, each endpoint of the contact and each own endpoint receives the same trust message by the server. Thus, a client needs to send the same trust message only once.
If not all endpoints of the contact should receive the trust message, the trust message MAY be sent to specific endpoints of the contact but for all own endpoints Message Carbons (XEP-0280)  MAY be used and vice versa. Even when a client does not yet have a contact, the client MAY use Message Carbons (XEP-0280)  for delivering a trust message to all own endpoints by sending it to the own bare JID. If then a client receives a trust message with its own full JID as the sender, it MAY discard that message directly without parsing the content.
Example: Alice's endpoint A1 authenticates the key of her endpoint A2. A1 sends the trust message for A2's key only once to all of Alice's and Bob's endpoints by using Message Carbons (XEP-0280) .
Attention: In that context, sending an encrypted trust message to all endpoints of a contact or to all own endpoints does not mean to encrypt it with the keys of all those endpoints. Instead, it only means that all of those endpoints should receive the trust message even if it is not encrypted for some of them and thereby not decryptable by those endpoints. Keep in mind that a trust message SHOULD only be encrypted for endpoints with authenticated keys.
This document requires no interaction with the Internet Assigned Numbers Authority (IANA).
This specification defines the following XMPP namespaces:
If the protocol defined in this specification undergoes a revision that is not fully backwards-compatible with an older version, the XMPP Registrar shall increment the protocol version number found at the end of the XML namespaces defined herein, as described in Section 4 of XEP-0053.
This document in other formats: XML PDF
This XMPP Extension Protocol is copyright © 1999 – 2020 by the XMPP Standards Foundation (XSF).
Permission is hereby granted, free of charge, to any person obtaining a copy of this specification (the "Specification"), to make use of the Specification without restriction, including without limitation the rights to implement the Specification in a software program, deploy the Specification in a network service, and copy, modify, merge, publish, translate, distribute, sublicense, or sell copies of the Specification, and to permit persons to whom the Specification is furnished to do so, subject to the condition that the foregoing copyright notice and this permission notice shall be included in all copies or substantial portions of the Specification. Unless separate permission is granted, modified works that are redistributed shall not contain misleading information regarding the authors, title, number, or publisher of the Specification, and shall not claim endorsement of the modified works by the authors, any organization or project to which the authors belong, or the XMPP Standards Foundation.
## NOTE WELL: This Specification is provided on an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, express or implied, including, without limitation, any warranties or conditions of TITLE, NON-INFRINGEMENT, MERCHANTABILITY, or FITNESS FOR A PARTICULAR PURPOSE. ##
In no event and under no legal theory, whether in tort (including negligence), contract, or otherwise, unless required by applicable law (such as deliberate and grossly negligent acts) or agreed to in writing, shall the XMPP Standards Foundation or any author of this Specification be liable for damages, including any direct, indirect, special, incidental, or consequential damages of any character arising from, out of, or in connection with the Specification or the implementation, deployment, or other use of the Specification (including but not limited to damages for loss of goodwill, work stoppage, computer failure or malfunction, or any and all other commercial damages or losses), even if the XMPP Standards Foundation or such author has been advised of the possibility of such damages.
This XMPP Extension Protocol has been contributed in full conformance with the XSF's Intellectual Property Rights Policy (a copy of which can be found at <https://xmpp.org/about/xsf/ipr-policy> or obtained by writing to XMPP Standards Foundation, P.O. Box 787, Parker, CO 80134 USA).
The Extensible Messaging and Presence Protocol (XMPP) is defined in the XMPP Core (RFC 6120) and XMPP IM (RFC 6121) specifications contributed by the XMPP Standards Foundation to the Internet Standards Process, which is managed by the Internet Engineering Task Force in accordance with RFC 2026. Any protocol defined in this document has been developed outside the Internet Standards Process and is to be understood as an extension to XMPP rather than as an evolution, development, or modification of XMPP itself.
The primary venue for discussion of XMPP Extension Protocols is the <firstname.lastname@example.org> discussion list.
Discussion on other xmpp.org discussion lists might also be appropriate; see <http://xmpp.org/about/discuss.shtml> for a complete list.
Errata can be sent to <email@example.com>.
The following requirements keywords as used in this document are to be interpreted as described in RFC 2119: "MUST", "SHALL", "REQUIRED"; "MUST NOT", "SHALL NOT"; "SHOULD", "RECOMMENDED"; "SHOULD NOT", "NOT RECOMMENDED"; "MAY", "OPTIONAL".
Note: Older versions of this specification might be available at http://xmpp.org/extensions/attic/
Improve explanations, descriptions and examples, introduce new attribute and complete all sections: