OpenSSL 3.5/3.6 TLS 1.3 Preferred KEX Tuple Bug (CVE-2026-2673)
CVE-2026-2673 Published on March 13, 2026

OpenSSL TLS 1.3 server may choose unexpected key agreement group
Issue summary: An OpenSSL TLS 1.3 server may fail to negotiate the expected preferred key exchange group when its key exchange group configuration includes the default by using the 'DEFAULT' keyword. Impact summary: A less preferred key exchange may be used even when a more preferred group is supported by both client and server, if the group was not included among the client's initial predicated keyshares. This will sometimes be the case with the new hybrid post-quantum groups, if the client chooses to defer their use until specifically requested by the server. If an OpenSSL TLS 1.3 server's configuration uses the 'DEFAULT' keyword to interpolate the built-in default group list into its own configuration, perhaps adding or removing specific elements, then an implementation defect causes the 'DEFAULT' list to lose its 'tuple' structure, and all server-supported groups were treated as a single sufficiently secure 'tuple', with the server not sending a Hello Retry Request (HRR) even when a group in a more preferred tuple was mutually supported. As a result, the client and server might fail to negotiate a mutually supported post-quantum key agreement group, such as 'X25519MLKEM768', if the client's configuration results in only 'classical' groups (such as 'X25519' being the only ones in the client's initial keyshare prediction). OpenSSL 3.5 and later support a new syntax for selecting the most preferred TLS 1.3 key agreement group on TLS servers. The old syntax had a single 'flat' list of groups, and treated all the supported groups as sufficiently secure. If any of the keyshares predicted by the client were supported by the server the most preferred among these was selected, even if other groups supported by the client, but not included in the list of predicted keyshares would have been more preferred, if included. The new syntax partitions the groups into distinct 'tuples' of roughly equivalent security. Within each tuple the most preferred group included among the client's predicted keyshares is chosen, but if the client supports a group from a more preferred tuple, but did not predict any corresponding keyshares, the server will ask the client to retry the ClientHello (by issuing a Hello Retry Request or HRR) with the most preferred mutually supported group. The above works as expected when the server's configuration uses the built-in default group list, or explicitly defines its own list by directly defining the various desired groups and group 'tuples'. No OpenSSL FIPS modules are affected by this issue, the code in question lies outside the FIPS boundary. OpenSSL 3.6 and 3.5 are vulnerable to this issue. OpenSSL 3.6 users should upgrade to OpenSSL 3.6.2 once it is released. OpenSSL 3.5 users should upgrade to OpenSSL 3.5.6 once it is released. OpenSSL 3.4, 3.3, 3.0, 1.0.2 and 1.1.1 are not affected by this issue.

Vendor Advisory NVD

Weakness Type

What is an Algorithm Downgrade Vulnerability?

A protocol or its implementation supports interaction between multiple actors and allows those actors to negotiate which algorithm should be used as a protection mechanism such as encryption or authentication, but it does not select the strongest algorithm that is available to both parties. When a security mechanism can be forced to downgrade to use a less secure algorithm, this can make it easier for attackers to compromise the software by exploiting weaker algorithm. The victim might not be aware that the less secure algorithm is being used. For example, if an attacker can force a communications channel to use cleartext instead of strongly-encrypted data, then the attacker could read the channel by sniffing, instead of going through extra effort of trying to decrypt the data using brute force techniques.

CVE-2026-2673 has been classified to as an Algorithm Downgrade vulnerability or weakness.


Products Associated with CVE-2026-2673

Want to know whenever a new CVE is published for OpenSSL? stack.watch will email you.

 

Affected Versions

OpenSSL: