OpenSSL CVE-2026-121481: is it really 9.8, and is there a fix?
Three published assessments of one base64 bug, two of them CVSS scores 6.1 points apart, and the patch is still an open pull request blocked on review. The affected range is narrower than any scanner will tell you: only the 4.0 and 4.1 lines.

No fixed OpenSSL release exists yet. CISA's enrichment scored CVE-2026-121481 at 9.8 on 19 November, OpenSSL's own record rates it Low, and the fix is still an unmerged pull request. Only the 4.0 and 4.1 lines are named, and nothing is exploited.
Why do the published scores disagree?
They come from three parties measuring different things. OpenSSL assigned the CVE itself on 18 November 2026 and rated it Low under its own security policy. On 19 November, CISA's ADP Vulnrichment attached a CVSS 3.1 base score of 9.8, severity CRITICAL, on the vector CVSS:3.1/AV:N/AC:L/PR:N/UI:N/S:U/C:H/I:H/A:H. Red Hat then published a third number, 3.7, on CVSS:3.1/AV:N/AC:H/PR:N/UI:N/S:U/C:N/I:N/A:L, saying it had "Lowered from prior Important" because OpenSSL rates the flaw Low with no known attack path. That is a spread of 6.1 points on one ten-point scale.
The two CWEs do not match either. OpenSSL files it as CWE-190, integer overflow or wraparound. CISA files the same flaw as CWE-122, heap-based buffer overflow. The 9.8 vector assumes low attack complexity and no privileges, while the record it is attached to says no attacker path has been found at all. We ran into the same shape of problem with the November Patch Tuesday counts, where two sources described one release with two numbers.
Which OpenSSL versions are affected?
The assigning CNA names two ranges and nothing else: 4.1.0 before 4.1.1, and 4.0.0 before 4.0.4. Every other version is marked unaffected by default. Red Hat states it the same way, saying only the "4.0/4.1 lines are affected upstream", and puts anything shipping OpenSSL 3.5 or earlier "outside the upstream affected range".
That is narrower than a scanner reporting a 9.8 against your whole estate will suggest. The 3.5 LTS line, which a lot of production still runs, is not in the range. NVD has not analysed the record: its status reads "Awaiting Analysis". The GitHub Advisory Database entry, GHSA-83xj-7xh4-4p7x, is marked "unreviewed".
Is there a patch to install yet?
No OpenSSL release carrying the fix has shipped yet. The download page lists three current releases: openssl-4.1.0.tar.gz from 23 October 2026, openssl-4.0.3.tar.gz from 29 September 2026, and openssl-3.5.9.tar.gz from 29 September 2026. The versions the CVE names as fixed, 4.1.1 and 4.0.4, do not appear on that page, on the newslog, or anywhere we could read. They are branch targets, not builds.
There is also no advisory. The newest file on OpenSSL's advisory index is dated 17 November 2026, the day before this CVE was published, and the vulnerabilities page does not list CVE-2026-121481 at all.
The fix is pull request 33652, opened by nmouha on 10 November 2026. Its current head guards EVP_EncodeBlock() with OPENSSL_assert() on negative and oversized lengths; the opening description, still displayed, describes an earlier approach of returning -1. Ten days on, the pull request is open, labeled among others "severity: important" and "hold: discussion", with the sidebar reading "Requested changes must be addressed". andrewkdinh approved on 19 November and jogme on 20 November. mattcaswell requested changes the same day, saying he is "strongly against" that use of OPENSSL_assert and that the library should return 0 instead. The body explains why the work is public at all: OpenSSL's security team asked for a public fix, and "the advisory is still pending".
The same thread carries the project's own answer to the 9.8. On 20 November mattcaswell wrote that handling a flaw in the open means it is "considered Low severity by the Project", and kroeckx noted that "The published severity is Low". On that account the Low rating follows from fixing the bug in the open, not from a judgment on the impact.
Can an attacker actually reach the overflow?
On OpenSSL's own account, not through anything it ships. The record says the bug needs "an input length close to INT_MAX", that TLS, QUIC and CMS "never encode buffers of that size", and that a caller needs "a buffer of about two gigabytes" in hand first. The exposed case is an application that takes a length from an attacker and passes it on "without its own bound check". There is no FIPS impact: the code sits outside the module boundary.
One claim cuts against that. The same record says its command line tools "cannot practically" be driven there, while Nicky Mouha, credited as both reporter and remediation developer, writes in the pull request that reaching it is "possible by passing a large" -proxy_pass to openssl s_client, with the password handed over in an environment variable because the command line would otherwise be too long. Both statements come from OpenSSL. Neither has been withdrawn. If you are writing the risk note, cite the one you can point at.
What do Red Hat, Ubuntu and Debian say?
All three have published, and all three say something different. Red Hat rates it LOW at 3.7 and prints a comparison table showing its 3.7 against cve.org's 9.8. Ubuntu sets priority Medium, and every openssl row reads "Needs evaluation". Debian's tracker has no per-release rows and one note: "check".
Ubuntu is the one to read closely if you run Node.js. Canonical marks nodejs on 22.04 LTS jammy as "Vulnerable", and a note explains why: that build "embeds OpenSSL 1.1.1m". That points outside the 4.0 and 4.1 range the CNA names, against Red Hat's statement that only those lines are affected. Canonical has not evaluated openssl itself, so neither position is settled. Embedded copies are where affected-version questions go wrong, as with the OpenSSH flaw in Alpine builds.
Is CVE-2026-121481 being exploited?
Nothing we read today says so. The CVE is not in CISA's Known Exploited Vulnerabilities catalog, and the record's KEV field is empty. CISA's own SSVC decision, timestamped 19 November 2026, records exploitation as "none", automatable as "yes" and technical impact as "total". OpenSSL's description is blunter still: no attacker path to the bug has been identified. EPSS puts the chance of exploitation activity at 0.14 percent.
Read together, the 9.8 is worst-case arithmetic on an impact nobody has shown how to trigger, published while the fix sits unmerged. The author compares the pattern to CVE-2021-23840.
What should IT teams do about this CVE today?
- Scope the ticket to OpenSSL 4.0.0 to 4.0.3 and 4.1.0 before you schedule anything. If you run 3.5 or earlier, Red Hat says you are outside the upstream range.
- Do not promise a patch date. There is no 4.1.1 or 4.0.4 release and no published advisory, so an upgrade window booked this week has nothing to install.
- Grep your own code for
EVP_EncodeBlock()calls that take a length from a request, a file header or an upload. That is the only path the record describes as exposed, and your own bound check closes it without waiting for OpenSSL. - Check embedded copies, not just the system package. Ubuntu's only "Vulnerable" verdict is a Node.js build on jammy that ships its own OpenSSL.
- If a scanner has raised this as critical, attach all three published ratings and the SSVC exploitation value of "none" to the risk acceptance. The 9.8 came from automated enrichment, not from OpenSSL.
- Watch pull request 33652 rather than the advisory page. The merge, not the CVE, is what will set the release date.
Sources
- CVE Record CVE-2026-121481, OpenSSL as assigning CNA, published 18 November 2026
- openssl/openssl pull request 33652, the fix for CVE-2026-121481
- OpenSSL downloads: latest release per branch
- OpenSSL security advisory index
- OpenSSL vulnerabilities page
- OpenSSL newslog: releases and security advisories
- Red Hat security advisory page for CVE-2026-121481
- Ubuntu security page for CVE-2026-121481
- Debian security tracker entry for CVE-2026-121481
- NVD API record for CVE-2026-121481
- GitHub Advisory Database GHSA-83xj-7xh4-4p7x
- CISA Known Exploited Vulnerabilities catalog