A base64 helper in OpenSSL has an integer-overflow bug scored CVSS 3.1 9.8. The vector reads network-reachable, low complexity, no privileges, no user interaction.
The same advisory says: “No path by which an attacker can trigger this has been identified.”
Both statements are in the record. Reading them together is the point.
The mechanics
EVP_EncodeBlock() encodes a buffer to base64. It sizes the output as 4*((len+2)/3)+1, and near INT_MAX that signed arithmetic overflows and the function writes past the caller’s buffer. The advisory tracks it as CWE-190, an integer overflow.
Reaching it needs a caller that already holds “a buffer of about two gigabytes”. That is a size the calling application must supply, not one a remote attacker can ask a library for.
What OpenSSL says is not exposed
The advisory names them: “OpenSSL’s TLS, QUIC and CMS implementations never encode buffers of that size, and the command line tools cannot practically be made to do so”.
So a TLS terminator, a QUIC server or the CLI is not the caller in question. The exposed case is narrower: an application that takes a length from somewhere untrusted and forwards it to EVP_EncodeBlock() without its own bound check. The advisory says so in those terms.
Its severity statement is blunt about the gap between the vector and the reachability: “The severity is Low because no attacker-controlled path to the overflow is known.” FIPS is not affected, because the code sits outside the module boundary.
Nothing has shipped the fix
Pull request openssl/openssl#33652, titled for this CVE, is open and unmerged. It was created on 10 November against master, and its labels include approval: review pending, hold: discussion, severity: important, and branches 4.0 and 4.1.
The newest release is 4.1.0 from 23 October, which predates the pull request by more than two weeks. Every other current release is older still. No released version contains this fix, and there is no published fixed-version table to point at.
The project’s vulnerability index did not list CVE-2026-121481 when checked. That is an observation about one check, not a forecast.
Why the pull request carries hold: discussion is not something these records explain, and this piece does not guess.
What to do with a 9.8 like this
Do not panic, and do not ignore it. The bug is real. The reason it is not urgent is reachability, not any doubt about the defect.
Your scanner will still shout. A CVE-keyed pipeline sees the 9.8 and the AV:N vector and ranks it beside genuinely remote bugs. This is why a vector is not an exploit.
The desk covered a Linux TCP use-after-free that carried the same 9.8, and that one had a documented chain to a root shell. The difference there was reachability, not the number.
If you maintain the calling application, the only mitigation the material supports is a bound check at the call site. Do that before the length reaches the function.
NVD lists the record as Awaiting Analysis, so the description is the vendor’s own. No exploitation in the wild is reported.
Sources: NVD, CVE-2026-121481, the GitHub advisory, pull request 33652, OpenSSL’s release list and the project’s vulnerability index, all read on 20 November 2026.

