Key takeaways

  • CVE-2026-121481 is a heap buffer overflow in OpenSSL's EVP_EncodeBlock() when the input length is close to INT_MAX.
  • CISA scored it 9.8 on NVD, while OpenSSL labelled its fix important and a reviewer called the reproducer synthetic.
  • As of November 20, 2026 the fix is an open pull request with no fixed release or OpenSSL advisory.

An OpenSSL vulnerability published on November 18, CVE-2026-121481, carries a CVSS score of 9.8 out of 10 in the US National Vulnerability Database. The bug is a heap buffer overflow in EVP_EncodeBlock(), OpenSSL's base64 encoding function. As of November 20 there is no fixed OpenSSL release and no official OpenSSL advisory: the fix is still an open pull request.

The score has made the bug the top "openssl vulnerability" story of the week, and it reached the Hacker News front page. The details suggest less danger than the number implies, but anyone who ships software linked against OpenSSL should know what is and isn't known.

What the bug is

The CVE record, filed by OpenSSL's own security team, says that calling EVP_EncodeBlock() with an input length close to INT_MAX, about 2 GB, "may overflow the calculation of the encoded output size and write past the end of the caller-supplied output buffer." In plain terms: hand the function an enormous input, and the size math wraps around, so it writes more than the buffer can hold.

The pull request also covers other invalid inputs: null buffers and negative lengths. Its current version stops the program with an assertion when it sees them, instead of returning an error, because reviewers pointed out that existing callers can't tell an error code from a valid result. The same check goes into EVP_DecodeBlock(), and a related length check in OpenSSL's HTTP client gets fixed.

Where the 9.8 comes from

OpenSSL did not assign the 9.8. It comes from CISA's enrichment of the record (the "ADP" score on NVD), which rates the bug as reachable over the network with no login. The same CISA entry records exploitation as "none."

OpenSSL itself labelled the pull request "severity: important," a label it uses for important bugs and for security fixes of any severity. A reviewer on the pull request wrote that the report's reproducer "is completely synthetic" and that it "probably should not be a CVE without evidence it really happens."

The author of the fix says it can be triggered from the command line by passing a very large -proxy_pass value to openssl s_client. That needs the attacker to control the input to begin with.

What to do now

  • There is nothing to install yet. OpenSSL's November 17 releases (4.1.1, 4.0.4, 3.5.10 and others) fixed three other, low-severity bugs, not this one.
  • The fix targets the master branch and the 4.0 and 4.1 lines. A maintainer noted that backports to older lines will need separate pull requests.
  • If your own code calls EVP_EncodeBlock() or EVP_DecodeBlock() with lengths that come from users, cap the input size well below 2 GB now. Code that never encodes inputs that large is not exposed to the overflow.
  • Watch OpenSSL's vulnerabilities page for the advisory, which will list the affected and fixed versions.

Security scanners that read NVD scores will flag every OpenSSL install as critical until fixed versions are out. Expect patch tickets before there is a patch to apply.

Sources

  1. CVE-2026-121481NIST National Vulnerability Database
  2. Fix EVP_EncodeBlock() buffer overflow (pull request #33652)OpenSSL
  3. GHSA-83xj-7xh4-4p7xGitHub Advisory Database
  4. VulnerabilitiesOpenSSL