&xep0060; and &xep0163; are widely used XMPP features. However, items are published in plain text, making them readable by server administrators or anybody having administrator level access on the pubsub/PEP services.

Although this is not a problem for data intended to be published publicly, there are many cases where it may be desirable to make the data inaccessible to anybody but the selected recipients, or in other words to end-to-end encrypt the data.

This specification aims to provide an easy way to e2e encrypt items of a pubsub node, and to share access with other entities. It is compatible with all existing pubsub/PEP features, which makes it possible to use securily features such a geolocation, private blogs, private calendar events, data forms, etc.

The design goals of this XEP are:

• work with basic standard pubsub/PEP service, no extra feature needed;
• be generic and work with any protocol using pubsub: once decrypted, the items should be parsable as if they were plain text;
• be able to share with other entities the means to decipher the elements;
• be able to easily share the means of decrypting items with new entities, even after the items have been published;
• be able to modify the means of decryption to make the new elements inaccessible to certain entities even if they have had access to the previous elements;

• e2ee: end-to-end encryption
• Shared Secret: a cryptographically strong random string used to symmetrically encrypt items
• Encrypted Node: pubsub node whose items are end-2-end encrypted using this protocol.

To encrypt a pubsub node, each item is symmetrically encrypted with a shared secret using OpenPGP symmetric encryption. The secret is shared with other entities using e2e encrypted messages with &xep0373;. If an entity's access is revoked or a shared secret may have been compromised, a new shared secret is generated and new items are encrypted with it.

The shared secret and revocation SHOULD be generated by node owner.

&MESSAGE; with encrypted <shared-secret/> or <revoke/> elements MUST be sent to everybody who must have access to the node. At least one of the encrypted payload SHOULD be encrypted for the sender themselves (either by sending the &MESSAGE; directly to the sender, or to decrypt a copy received with &xep0280;), so other devices of the sender can get the shared secret too.

When &xep0470; are used, the attachment node SHOULD have the same access model (i.e. whitelist as recommended in Security Considerations) as the encrypted node, with authorized entities synchronized (this should be done automatically for fully compliant services). The items published to attachment node itself MUST be encrypted using this protocol. Due to validity check of attachment items, the encrypted element MUST be a child of a XEP-0470's <attachments/> element, instead of being the <item/> payload as usual. Note that this will prevent the summary feature to work with encrypted elements, and the end-user client will have to do the summary itself, this is an inevitable trade-off when using e2ee.

The "pubsub#type" of an encrypted item is always "urn:xmpp:openpgp:pubsub:0", thus no information is leaked on the content of the node, and all encrypted nodes can easily be retrieved by using &xep0462;.

Note that encrypted items MAY be mixed with plain text items: for instance if a blog is public, but some of its items are private. However the proper handling of this use case is out of scope of this specification.

This specification uses &xep0373; independently of &xep0374;, implementations SHOULD avoid double encryption (e.g. by excluding <openpgp/> element qualified by 'urn:xmpp:openpgp:0' namespace from encryption by XEP-0374 implementation), and MUST ensure that element is correctly decrypted. Double encryption should be checked, and the <openpgp/> element MUST be decrypted even if OXIM (XEP-0374) is not being used.

If a client supports the protocol specified in this XEP, it MUST advertise it by including the "urn:xmpp:openpgp:pubsub:0" discovery feature in response to a &xep0030; information request:

• When receiving a <shared-secret/> or a <revoke/> element, the receiving client MUST ensure that the signing sender is the same as the one for all known shared secrets of this pubsub node (usually a node owner). This prevents a malicious actor for misleading the client into using a non-legitimate secret.
• To limit the surface of attack, the access model of an encrypted node should be set to "whitelist" and only people having the shared key should be allowed to retrieve encrypted items.
• If the shared key is compromised, or a user access is revoked, the key MUST be rotated. However, only new items are encrypted with the new key, any previous item should be considered as compromised too.
• Sometimes client may use metadata to construct item ID, this is notably the case for some &xep0277; implementation, as the resulting item ID is used to generate user friendly URLs. To avoid metadata leakage, clients SHOULD NOT derivate the item ID from any data of the item when pubsub encryption is used.
• Some protocols define prefixes for nodes, this is for instance the case for XEP-0277 Comments. If the prefix is not essential to make the feature work (which is the case for XEP-0277 comments), the client SHOULD NOT use the usual prefix to avoid leaking informations on the content of the node. The prefix defined in &xep0470; MUST be used as usually, as it is essential to retrieve the attachment node, and it's generic and thus doesn't leak information on the node content.
• To decrypt archives and future items, clients most probably cache plain text version of items, and shared secret. If one device with access to an encrypted node is compromised, all items from the node should be considered compromised, and shared secret should be rotated.
• For the same reason, client SHOULD encrypt data at rest (even if the device is stolen, data should not be accessible without some cryptographic key).
• Note that only the shared key sender is authenticated (by &xep0373;'s <signcrypt/>), not the items publishers, meaning that encrypted items could be published by anybody in possession of the shared secret. Pubsub items authentication will be treated in a separated XEP.
• The shared secret should be long enough to avoid brute force attacks. A secret of at least 32 characters is recommended.

TODO

TODO

TODO

Thanks to Tim Henkes for his advices and NLNet foundation/NGI0 Discovery for funding.