OpenSSL Security Advisory
41 views
Skip to first unread message
Tomas Mraz
Mar 13, 2026, 9:23:37 AM (11 days ago) Mar 13
to openssl-project, openssl-users
OpenSSL Security Advisory [13th March 2026]
===========================================
OpenSSL TLS 1.3 server may choose unexpected key agreement group (CVE-2026-2673)
================================================================================
Severity: Low
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.
Due to the low severity of this issue we are not issuing new releases of
OpenSSL at this time. The fix will be included in the next release of 3.6
and 3.5 branches, once it becomes available. The fix is also available in commit
2157c9d8 (for 3.6) and commit 85977e01 (for 3.5) in the OpenSSL git repository.
This issue was internally reported on the 16th of February 2026 by Viktor
Dukhovni. The fix was developed by Viktor Dukhovni.
General Advisory Notes
======================
URL for this Security Advisory:
https://openssl-library.org/news/secadv/20260313.txt
Note: the online version of the advisory may be updated with additional details
over time.
For details of OpenSSL severity classifications please see:
https://openssl-library.org/policies/general/security-policy/
===========================================
OpenSSL TLS 1.3 server may choose unexpected key agreement group (CVE-2026-2673)
================================================================================
Severity: Low
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.
Due to the low severity of this issue we are not issuing new releases of
OpenSSL at this time. The fix will be included in the next release of 3.6
and 3.5 branches, once it becomes available. The fix is also available in commit
2157c9d8 (for 3.6) and commit 85977e01 (for 3.5) in the OpenSSL git repository.
This issue was internally reported on the 16th of February 2026 by Viktor
Dukhovni. The fix was developed by Viktor Dukhovni.
General Advisory Notes
======================
URL for this Security Advisory:
https://openssl-library.org/news/secadv/20260313.txt
Note: the online version of the advisory may be updated with additional details
over time.
For details of OpenSSL severity classifications please see:
https://openssl-library.org/policies/general/security-policy/
Dimitri Ledkov
Mar 18, 2026, 3:49:05 PM (6 days ago) Mar 18
to Tomas Mraz, openssl-project, openssl-users
The NVD has updated the CVE https://nvd.nist.gov/vuln/detail/CVE-2026-2673
The assigned score is CVSS:3.1/AV:N/AC:L/PR:N/UI:N/S:U/C:H/I:N/A:N
which is classed as High by multiple security scanners.
Due to lack of upstream tags and tarballs, I have prepared backports
of the fix onto stable releases 3.5.5+1 and 3.6.1+1 and published that
at: https://github.com/openssl-stable/openssl-stable/releases
Any users can reuse the git cherry-picks, rebased and conflicts
resolved, or download immutable releases from there.
I again call for OpenSSL upstream to reconsider and tag releases for
any CVE. It is always best to branch last point release, commit the
fix, tag it, and merge into the updates branch. This way, tagged
releases require no additional work. Git describe can then correctly
identify the builds before and after the fix on the updates branch as
those with or without the fix.
The OpenSSL assessment and classification of CVEs may not align with
classification determined by the regulators. Hence all CVE patches
should follow the same process once they are made public.
Regards,
Dimitri.
> --
> You received this message because you are subscribed to the Google Groups "openssl-project" group.
> To unsubscribe from this group and stop receiving emails from it, send an email to openssl-proje...@openssl.org.
> To view this discussion visit https://groups.google.com/a/openssl.org/d/msgid/openssl-project/10107223b33073994e235de61f20bde963d8fe0b.camel%40openssl.foundation.
The assigned score is CVSS:3.1/AV:N/AC:L/PR:N/UI:N/S:U/C:H/I:N/A:N
which is classed as High by multiple security scanners.
Due to lack of upstream tags and tarballs, I have prepared backports
of the fix onto stable releases 3.5.5+1 and 3.6.1+1 and published that
at: https://github.com/openssl-stable/openssl-stable/releases
Any users can reuse the git cherry-picks, rebased and conflicts
resolved, or download immutable releases from there.
I again call for OpenSSL upstream to reconsider and tag releases for
any CVE. It is always best to branch last point release, commit the
fix, tag it, and merge into the updates branch. This way, tagged
releases require no additional work. Git describe can then correctly
identify the builds before and after the fix on the updates branch as
those with or without the fix.
The OpenSSL assessment and classification of CVEs may not align with
classification determined by the regulators. Hence all CVE patches
should follow the same process once they are made public.
Regards,
Dimitri.
> You received this message because you are subscribed to the Google Groups "openssl-project" group.
> To unsubscribe from this group and stop receiving emails from it, send an email to openssl-proje...@openssl.org.
> To view this discussion visit https://groups.google.com/a/openssl.org/d/msgid/openssl-project/10107223b33073994e235de61f20bde963d8fe0b.camel%40openssl.foundation.
Viktor Dukhovni
Mar 18, 2026, 10:26:53 PM (6 days ago) Mar 18
to openss...@openssl.org, openssl-project
On Wed, Mar 18, 2026 at 03:48:21PM -0400, 'Dimitri Ledkov' via openssl-users wrote:
> The NVD has updated the CVE https://nvd.nist.gov/vuln/detail/CVE-2026-2673
>
> The assigned score is CVSS:3.1/AV:N/AC:L/PR:N/UI:N/S:U/C:H/I:N/A:N
> which is classed as High by multiple security scanners.
We're not responsible for the folly of others. This issue affects
> The NVD has updated the CVE https://nvd.nist.gov/vuln/detail/CVE-2026-2673
>
> The assigned score is CVSS:3.1/AV:N/AC:L/PR:N/UI:N/S:U/C:H/I:N/A:N
> which is classed as High by multiple security scanners.
basically nobody today and for some time to come. A CVE was assigned
only to make sure that distros don't skip integrating the fix into their
stable releases when they choose which security fixes to backport to
their stable OS updates.
Eventually, as PQC-capable clients and servers become more common, and
as servers start to use the "DEFAULT" keyword in their supported groups
configurations prefixing or suffixing it with minor tweaks, the bug fix
will be desirable.
For now the only affected software that comes to mind is bleeding-edge
Postfix 3.11 builds (released 2026-03-05!) on systems with bleeding-edge
OpenSSL runtimes (3.5 or later, released 2025-04-08), when communicating
with similar systems, and only to the extent that HNDL (harvest now
decrypt later) is a real concern (exceptionally high-value target
emailing highly confidential data, protected only by SMTP STARTTLS).
This is because Postfix 3.11 is willing to accept higher latency in
order to avoid occasional problems with middleboxes that choke on larger
PQC TLS Client Hellos, and so by default has (essentially):
+#if OPENSSL_VERSION_PREREQ(3,5)
+#define DEF_TLS_EECDH_AUTO "?X25519MLKEM768:DEFAULT"
+#define DEF_TLS_FFDHE_AUTO ""
+#else
... settings for older OpenSSL
+#endif
When both the Postfix client and server are running the days old 3.11,
this default configuration setting trips over the bug, with servers
accepting the client's X25519 keyshare, rather sending an HRR for
X25519MLKEM768.
> Due to lack of upstream tags and tarballs, I have prepared backports
> of the fix onto stable releases 3.5.5+1 and 3.6.1+1 and published that
> at: https://github.com/openssl-stable/openssl-stable/releases
falling.
--
Viktor. 🇺🇦 Слава Україні!
Tomas Mraz
Mar 19, 2026, 4:30:26 AM (6 days ago) Mar 19
to Dimitri Ledkov, openssl-project, openssl-users
The NVD classification for this basically non-issue is just hilarious.
I have no further comments.
Tomas Mraz, CTO, OpenSSL Foundation
--
Tomáš Mráz, Chief Technology Officer, OpenSSL Foundation
We need your support! Help us protect digital privacy… everywhere.
https://openssl.foundation/donate/ways-to-give
I have no further comments.
Tomas Mraz, CTO, OpenSSL Foundation
Tomáš Mráz, Chief Technology Officer, OpenSSL Foundation
We need your support! Help us protect digital privacy… everywhere.
https://openssl.foundation/donate/ways-to-give
Reply all
Reply to author
Forward
0 new messages