The Federation API specification specifies that events should be validated via the signature verification algorithm, but does not specify how the keys for that check should be obtained and validated.
In practice, the implementation has been as follows. The receiving server
first requests a copy of the key via the GET /_matrix/key/v2/server/
API
directly from the server which created the signature, or via the POST /_matrix/key/v2/query API
from a trusted key server. Once such a key is obtained, it is then cached
forever. No check is made on the valid_until_ts field, and
minimum_valid_until_ts is set to zero for calls to POST /_matrix/key/v2/query.
This is highly unsatisfactory, as it means that, should a key be compromised, then an attacker can spoof arbitrary events claiming to be from the compromised server forever, since there is no revocation mechanism.
This MSC proposes to enforce the valid_until_ts property when validating
event signatures. In particular, the server must ensure that it has a copy of
the key with a valid_until_ts at least as large as the origin_server_ts of
the event being validated. If it does not have such a copy, it must try to
obtain one via the GET /_matrix/key/v2/server/ or POST /_matrix/key/v2/query APIs. For the latter, it must set
minimum_valid_until_ts to prompt the notary server to attempt to refresh the
key if appropriate.
Since this changes the rules used to validate events, it will be introduced with a new room version. This will reduce the risk of divergence between servers in a room due to some servers accepting events which others reject.
This MSC also proposes that the current situation - where valid_until_ts is
ignored - be formalised for the existing room versions v1-v4, rather than be
left as implementation-specific behaviour.