Java4Shell Scores 9.2 From Its Finder and 8.1 From OpenJDK, and Needs an AVX-512 CPU
Security / news
Java4Shell Scores 9.2 From Its Finder and 8.1 From OpenJDK, and Needs an AVX-512 CPU
Calif says a 65-character Base64 string overwrites 62 bytes of Java heap and reaches code execution in Keycloak, Jenkins and Jira. A second security firm says the source shows only a crash.

An unauthenticated remote attacker can overwrite 62 bytes of heap memory in OpenJDK 17, 21, 25 and 27 by sending a malformed Base64 string, provided the server runs on an x86-64 CPU with AVX-512 VBMI. Calif.io disclosed the flaw on Nov. 17 as Java4Shell (CVE-2026-114938) and says it reaches remote code execution. The fix is a vendor update; two JVM flags are a stopgap that no vendor has validated.

Two scores for CVE-2026-114938
The score overstates or understates depending on whose vector you read. Calif's write-up rates the flaw 9.2 under CVSS 4.0. The OpenJDK advisory, last updated at 17:44 UTC the same day, lists it under hotspot/compiler at 8.1 under CVSS 3.1, with the vector AV:N/AC:H/PR:N/UI:N/S:U/C:H/I:H/A:H.
The difference is attack complexity. OpenJDK scored it high, which fits Calif's own admission that every target needs a bespoke profile and that the attack is sensitive to network round-trip time and jitter.
| Release line | Affected per OpenJDK matrix | Fixed build confirmed |
|---|---|---|
| JDK 8 | No | n/a |
| JDK 11 | No | n/a |
| JDK 17 | Yes | Not by OpenJDK |
| JDK 21, 25, 27 | Yes | Not by OpenJDK |
Calif's page lists Oracle build 8u511b33-perf among the Java SE versions Oracle called vulnerable, while the OpenJDK matrix leaves the JDK 8 column empty. The two pages do not reconcile.
What the bug does
The flaw sits in the hand-written assembly behind java.util.Base64, which uses AVX-512 VBMI instructions to decode 64 bytes at a time. For a 65-character input (64 A characters plus =; 128 plus = on older JDK 17 builds), Java allocates a 50-byte result array. The routine then computes a negative output length for the stray =, the write mask becomes all ones, and 64 bytes land in the heap starting at offset 48.
Java rejects the input afterwards, but the neighbouring objects are already damaged. ARM64 servers and x86 CPUs without VBMI never take this path.
Does it reach code execution?
Calif says yes. It lists pre-authentication code execution in default configurations of Keycloak, Atlassian Jira, Atlassian Confluence, JFrog Artifactory, Jenkins and PingFederate, and post-authentication execution in Kafka, Solr and Elasticsearch. The product exploit chains are withheld until, in its words, everyone has had a chance to upgrade.
Equixly, a security vendor, disagrees. Its post says the written bytes are "fixed, not chosen by the attacker" and that "The source does not support that yet." It calls denial of service the realistic impact.
Calif's answer is in its write-up: only the landing position is controllable, so it overwrites a buffer's length field, then uses that enlarged buffer to corrupt a neighbour with chosen bytes. A wrong guess can crash the JVM, it says, though often the attacker can retry.
Who found it and how fast
Quang Luong of Calif found the bug by pointing Codex at the JIT compiler, intrinsics and cryptography code. Calif says discovery took an hour, the first working exploit two more hours, and the token bill came in under $200. Those are Calif's figures and nobody independent has repeated them. The OpenJDK advisory credits Luong among nine acknowledged reporters.
The case for taking it seriously rests on the same write-up. It includes a 4096-byte-array proof of concept that prints REPRODUCED when exactly 62 bytes change. The case for calm is that the working chains are private.
What to do on Nov. 17
Apply the Oracle or distributor update, then restart every JVM. Orca Security's analysis says Temurin, Corretto and Zulu release on their own schedules and that Oracle's notes did not name the CVE when it wrote; the OpenJDK advisory does name it.
Until patched, -XX:-UseBASE64Intrinsics disables the fast path. -XX:UseAVX=2 avoids AVX-512 but, per Calif, stops an ARM JVM from starting. Equixly notes no vendor has confirmed either flag for this CVE.
The next marker is Calif's release of the product chains, which it has tied to upgrade progress and not a date. Related: Google's account of an exploit agent and the FBI's Artex warning cover the same AI-assisted speed-up from the attacker side.
Sources
More in Security
- 01CISA Gave 9 of 16 November KEV Entries a Three-Day Deadline With Forensic Triage, and 7 Got 14 DaysThe catalog's own due-date and forensicTriage fields show two tiers under BOD 26-04. Sorting November's additions by tier shows which exploited bugs CISA treats as already-compromised-until-proven-otherwise.
- 02Project Zero Cuts Its Disclosure Deadline to 60 Days on Feb. 2, 2027, Then 30 in 2028Google's bug-hunting team is dropping the 30-day patch soak period and counting every deadline from the day it reports, citing AI-assisted exploitation.
- 03Google Finds exploit_agent Diffing Patches and Firing Exploits With No Operator in the LoopGTIG's November 16 tracker reports the first fully automated discovery-to-exploit pipeline it has seen, found on an exposed staging server, plus 22 zero-days in August.
- 04Mandiant: LLMjacking Runs Up $100,000 a Day on Stolen Cloud Keys, and Logs Are Off by DefaultGoogle's Nov. 2 report says single-token test calls evade volume alerts, and it describes the LiteLLM bug differently from BleepingComputer.