Multi-RecipientInfo Bleichenbacher Oracle in CMS_decrypt() and PKCS7_decrypt()

Multi-RecipientInfo Bleichenbacher Oracle in CMS_decrypt() and PKCS7_decrypt() (CVE-2026-42768)
Severity: Low

Issue summary: The CMS_decrypt and PKCS7_decrypt functions are vulnerable to
Bleichenbacher-style attack when an attacker is able to provide the CMS or
S/MIME messages and observe the error code and/or decryption output.

Impact summary: The Bleichenbacher-style attack allows an attacker to use the
victim's vulnerable application as a way to decrypt or sign messages with the
victim's private RSA key.

The attack is possible in 2 variants.

The decryption API (CMS_decrypt(), PKCS7_decrypt()) is used without
providing the recipient certificate. In this case OpenSSL iterates over every
KeyTransRecipientInfo (KTRI) without stopping at the first success.
An attacker who authors a message with two KTRI entries — the first one
wrapping a real CEK under the victim's public key, the second with an
arbitrary probe ciphertext — obtains opportunity to iterate the 2nd KTRI to
get a valid PKCS#1 v1.5 padding if the error code of the application is
available.

That is a Bleichenbacher oracle (Bleichenbacher, CRYPTO '98): an
adaptive-chosen-ciphertext side channel from which the attacker decrypts any
RSA ciphertext to the victim's key or forges any PKCS#1 v1.5 signature under
it.

When the decryption API (CMS_decrypt(), PKCS7_decrypt()) is provided with
the recipient certificate, and the recipient is not found, a random
key is substituted.
An attacker who authors a message and is able to compare both error code and
the result of the decryption, can mount a Bleichenbacher oracle.

We are not aware of any applications that provide a remote attacker
an opportunity to mount an attack described in these scenarios. We consider
the existence of such application very unlikely, and for this reason this
CVE has been evaluated as Low severity.

To avoid these attacks, when RSA PKCS#1 v1.5 Key Transport is in use, the
invoked EVP_PKEY_decrypt() will use the implicit rejection mechanism described
in draft-irtf-cfrg-rsa-guidance. In previous OpenSSL releases the implicit
rejection was explicitly disabled.

The implicit rejection mechanism always returns a plaintext value,
the symmetric key. This result is deterministic for the ciphertext and the
private key. The length of the decryption result can happen to match the
length of the key of the symmetric cipher that was used for the content
encryption. When a certificate is not provided, the last RecipientInfo
producing a key that looks valid will be used. It may cause getting garbage
content on decryption. As a proper way to deal with this a recipient
certificate has to be provided to identify the particular RecipientInfo for
decryption.

The FIPS modules in 4.0, 3.6, 3.5, and 3.4 are not affected by this issue, as
CMS and S/MIME processing happens outside the OpenSSL FIPS module boundary.

OpenSSL 4.0, 3.6, 3.5, 3.4 are vulnerable to this issue.

OpenSSL 3.0, 1.1.1, and 1.0.2 do not implement RSA implicit rejection.
These branches are affected by variants of Bleichenbacher/Marvin attack, but
the proper fix would require a breaking change of RSA decryption behavior,
which is unfeasible for stable branches.

OpenSSL 4.0 users should upgrade to OpenSSL 4.0.1.
OpenSSL 3.6 users should upgrade to OpenSSL 3.6.2.
OpenSSL 3.5 users should upgrade to OpenSSL 3.5.6.
OpenSSL 3.4 users should upgrade to OpenSSL 3.4.5.

This issue was reported by Alex Gaynor (Anthropic) on 16th April 2026.
The fix was developed by Dmitry Belyavskiy (Red Hat) and Alicja Kario (Red
Hat) based on reporter's recommendations.