OpenSSL's CVE-2026-121481 Scores 9.8, but a Reviewer Calls the Reproducer Synthetic
Security / news
OpenSSL's CVE-2026-121481 Scores 9.8, but a Reviewer Calls the Reproducer Synthetic
The fix for the EVP_EncodeBlock() overflow is unmerged, the fixed releases 4.0.4 and 4.1.1 are unreleased, and the pull request label says "important", not Critical.
An unauthenticated network attacker could, on paper, overflow a heap buffer through OpenSSL's EVP_EncodeBlock() base64 function (CVE-2026-121481, CVSS 9.8), but only in an application that passes it an input length close to 2,147,483,647 bytes. The fix has not been merged. On November 13 a reviewer on the project argued in the pull request that the report's reproducer "is completely synthetic" and that the issue "probably should not be a CVE without evidence it really happens."
The advisory entry went live on November 18. The patch itself is still pull request #33652, opened on November 10 by a contributor using the handle nmouha. The OpenSSL security response team asked for the fix to be made in public while its own advisory stays pending, according to the pull request description.
What the record says and what it leaves out
The National Vulnerability Database record, published November 18 with the status "Awaiting Analysis", describes an overflow in the calculation of the encoded output size that can write past the end of the caller's buffer. It lists OpenSSL 4.1.0 before 4.1.1 and 4.0.0 before 4.0.4 as affected. No source fetched shows either of those fixed releases as published.
The 9.8 score carries the vector AV:N/AC:L/PR:N/UI:N, meaning a network attacker with no privileges and no user interaction. The score is listed as a secondary source entry, not as an analysis by NVD staff. The GitHub Advisory Database entry rates it Critical, marks it unreviewed, lists no patched version and gives an exploit prediction score of 0.14%, which sits at the 3rd percentile.
| Source | Rating | Fixed version listed |
|---|---|---|
| NVD record, Nov. 18 | CVSS 9.8 | None yet |
| GitHub Advisory Database | Critical, unreviewed | None yet |
| OpenSSL pull request label | "severity: important" | Open, not merged |
The CISA coordinator entry attached to the NVD record lists exploitation as "none", automatable as "yes" and technical impact as "total". That reading says an attacker could script it if a vulnerable code path were reachable, and that nobody has been seen doing so.
The precondition that decides the real risk
The score assumes a network attacker can reach the function with an enormous length. The pull request description gives the one route its author confirmed. Passing a very large -proxy_pass value to openssl s_client triggers it, and the author wrote that GOST accepts such large passwords by default. The value has to come through an environment variable, since a command line that long would be refused.
That is a local command-line path, and the victim is the person running the tool. It is not a remote attacker sending a request to a server. The score overstates the exposure of a typical server and understates nothing the pull request can show. An application that feeds attacker-controlled data of about 2 GiB into the encoder, through a path that never checks length, would be exposed. The sources do not identify one.
The project's own label on the pull request reads "severity: important", which disagrees with the Critical rating in the two public databases. The pull request also carries labels for the master, 4.0 and 4.1 branches.
How the fix changed in review
The first version returned -1 for a NULL buffer, a negative length or an encoded size that does not fit in an int. On November 12 the reviewer t8m said he was "really worried" that third-party callers might cast the return value to size_t without checking it.
On November 13 the reviewer bob-beck wrote that returning an error from these functions was "the incorrect fix", because 0 is already a valid success value and the project's STYLE.md says not to add error cases that look like success. He proposed documenting that negative lengths are not accepted. The contributor moved to OPENSSL_assert(), then dropped it again on November 17 after further review.
A follow-up, issue #33812, was opened on November 17 to move the overflow check into the internal functions. The reviewer andrewkdinh approved the pull request on November 19 and wrote that backports to the 4.0 and 4.1 branches will need separate pull requests.
What operators should do
Nothing ships yet, so there is no version to install. Check whether any in-house code calls EVP_EncodeBlock() with a length taken from untrusted input, and cap that length well below 2 GiB. The site's earlier report on an OpenJDK base64 flaw covers a different project, and the CISA exploited-vulnerabilities roundup of November 18 shows what the catalogue lists as exploited. The OpenSSL team has not said when 4.0.4 and 4.1.1 will be released.
Sources
More in Security
- 01Windows Server 2028 Preview Hardens Server Core by Default in Build 29706, but NTLM Is Still OpenHundreds of settings are on from first boot, Microsoft warns Server 2025 workloads may break, and the 33 percent of servers still on 2019 are untouched.
- 02VicOne's 28 Robotics Zero-Days Come From a Sponsored Article With No CVE IDs and No Named ProjectsThe Tokyo security firm's count ran as paid content on IEEE Spectrum and The Robot Report. An independent March paper counted 38 flaws in three named consumer robots.
- 03Synology's Third Critical DSM Advisory in 62 Days Hits OAuth, and the Last Patch Does Not Cover ItCVE-2026-120382 scores 9.8 and needs no login. DSM 7.4 and 7.3 builds ending -2 and -6 fix it; 7.2 and 7.1 have none yet.
- 04Codex's Sandbox Uses Seatbelt, bubblewrap and MXC, and Lets Commands Read Broadly by DefaultOpenAI's Nov. 18 engineering post says writes are confined to the workspace and the network is off, while reads stay wide open.