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.
In the terminology of &xep0384;, that is a <em>device</em>.
To cover also the possibility for using multiple endpoints on the same physical device and via the same client instance, the general term <em>endpoint</em> is used.
Trust messages can be used in conjunction with an end-to-end encryption protocol such as &xep0373; or &xep0384; to automatically or semi-automatically establish secure channels protected against active attacks.
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.
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.
</p>
</section3>
<section3topic='Confidential Communication of Trust'anchor='why-trust-messages-advantages-with-encryption-confidential-communication-of-trust'>
<p>
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.
</p>
<p>
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.
</p>
<p>
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.
</p>
<p>
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.
</p>
<p>
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.
</p>
<p>
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.
MUST contain exactly one <![CDATA[<trust-message/>]]> direct child element which
<ul>
<li>
MUST be signed in a way to ensure its authenticity and integrity.
</li>
<li>
SHOULD be encrypted to ensure its confidentiality.
</li>
<li>
MUST have an <em>xmlns</em> attribute specifying its namespace <em>&ns;</em>.
</li>
<li>
MUST have a <em>usage</em> attribute specifying the namespace of the protocol which uses the trust message for a specific purpose.
</li>
<li>
MUST have an <em>encryption</em> attribute specifying the namespace of the encryption protocol for which the keys are used.
</li>
<li>
MUST contain at least one <![CDATA[<key-owner/>]]> direct child element which
<ul>
<li>
MUST have a <em>JID</em> attribute specifying the owner of the keys.
</li>
<li>
MUST contain at least one <![CDATA[<trust/>]]> or <![CDATA[<distrust/>]]> direct child element which indicates the trust respectively distrust in a key.
Each <![CDATA[<trust/>]]> and <![CDATA[<distrust/>]]> element MUST contain exactly one key identifier.
</li>
</ul>
</li>
</ul>
</li>
<li>
SHOULD have a <![CDATA[type='chat']]> attribute which is needed to deliver the trust message to all endpoints (see <linkurl='https://xmpp.org/extensions/xep-0280.html#recommended-rules'>XEP-0280: Message Carbons</link>).
</li>
<li>
SHOULD contain a <![CDATA[<store xmlns='urn:xmpp:hints'/>]]> direct child element which is needed to deliver the trust message to each offline endpoint after it went online (see <linkurl='https://xmpp.org/extensions/xep-0313.html#business-storeret-user-archives'>XEP-0313: Message Archive Management</link> and <linkurl='https://xmpp.org/extensions/xep-0334.html#sect-idm45856619663120'>XEP-0334: Message Processing Hints</link>).
The last two points are needed to achieve their mentioned goals because a trust message does not contain a <![CDATA[<body>]]> element which would automatically lead to the desired result.
In the following example, two &xep0384; 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 <![CDATA[<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 <![CDATA[<trust-message/>]]> element has to be used by each existing encryption protocol is out of scope.
&xep0420; specifies a common method for encrypting arbitrary elements which can be used by different encryption protocols.
When using an encryption protocol such as &xep0384; which uses &xep0420; (SCE), the SCE <![CDATA[<payload/>]]> element MUST contain the <![CDATA[<trust-message/>]]> element as a direct child.
A trust message SCE <![CDATA[<content/>]]> element
</p>
<ul>
<li>
MUST contain an <![CDATA[<rpad/>]]> affix element.
This prevents an attacker from finding out the trust message's content by its length and distinguishing a trust message from another kind of SCE message.
</li>
<li>
MUST contain a <![CDATA[<time/>]]> affix element.
This prevents an attacker from delivering trust messages in the wrong order or delivering a former trust message again.
If the trust messages are used by the recipient for making trust decisions, it protects the recipient from setting the opposite of the intended trust state by applying trust changes in the wrong order or reapplying a former change.
</li>
<li>
SHOULD contain a <![CDATA[<from/>]]> affix element.
This prevents an attacker from spoofing the sender.
</li>
<li>
SHOULD contain a <![CDATA[<to/>]]> affix element.
This prevents an attacker from spoofing the recipient.
</li>
</ul>
</section3>
<examplecaption='SCE Content Element Used for a Trust Message from Alice's Endpoint to Carol'><