OpenJDK's AVX-512 Base64 decoder can corrupt the heap on malformed input

Someone reported java4shell bug in September 2026

OpenJDK's AVX-512 Base64 decoder can corrupt the heap on malformed input

A pull request against OpenJDK's JDK repository describes a bug in the x86_64 AVX-512 VBMI Base64 decode stub: when a short padded tail leaves fewer than 4 bytes, the padding path drives output_size below zero, producing a wrong mask and storing past the logical output length into the Java heap. Valid inputs always have lengths that are multiples of four, so only invalid input triggers it. The proposed fix returns early to Java when the remaining length is under 4. The PR was later closed without merging.

The stub stores past the logical output length before returning to Java, corrupting the heap.
  1. black3r

    The very first referenced issue in that PR discussion sheds a bit more light on this: https://github.com/adoptium/adoptium-support/issues/1480

    This PR's author says there on October 1st: "Got an email from Oracle asking me to NOT proceed with a PR until receiving approval." and then on October 28th: "It's taken over by Oracle. Nothing required from us."

  2. wood_spirit

    So the implication is that the LLM that assisted the Calif researchers might have read this previous bug report and not independently discovered it?

  3. darkamaul

    We will never know, but there is a chance that the Calif's researcher model decided to look at the open and closed issues / PRs of the repo.

    Nonetheless, being able to understand the criticality of a finding, and how to exploit it are other skills, that still require supervision.

  4. rzmmm

    From califio: "I focused the audit on the JIT compiler, intrinsics, widely accessible APIs, and cryptography. Well, more precisely, I told Codex to look at those areas like any serious security researcher in 2026."

    Maybe simply a coincidence, but I would put my bets that the first person with the PR was also using Codex, and the model was exposed to the details of the bug.

  5. InsidiousWombat

    Nice proof that the auto-close bots are literal cancer on project that make stuff worse.

    It was literal bug with provided fix. All it needed is someone to have a proper look.

    But someone clowned a bot under a visionary thought "if nobody looked it must not be important" and here we are

More from this day

2026-11-19