cd /news/ai-safety/oracle-and-openjdk-patch-java4shell-… · home › topics › ai-safety › article
[ARTICLE · art-171926] src=startupfortune.com ↗ pub= topic=ai-safety verified=true sentiment=↓ negative

Oracle and OpenJDK patch Java4Shell, a critical Java flaw found with AI

Oracle and the OpenJDK project released fixes on November 17 for CVE-2026-114938, a memory corruption flaw nicknamed Java4Shell in the JVM's fast Base64 decoder on x86-64 processors, with patched builds OpenJDK 17.0.21.1, 21.0.13.1, 25.0.5.1 and 27.0.1.1 plus Java SE 8u513. Oracle rates the flaw 8.1 on CVSS 3.1 while the US security startup Calif, which found it, scores it 9.2 and says it demonstrated unauthenticated code execution against Keycloak, Atlassian Jira and Confluence, JFrog Artifactory, Jenkins and PingFederate. CERT-EU says exploitation requires an x86-64 CPU with AVX-512 extensions (Intel Xeon Ice Lake onward or AMD Zen 4 and later) and attacker-controlled Base64 data reaching the decoder, so ARM64 machines including AWS Graviton are unaffected; no successful attacks have been reported.

recommendation

Restart every JVM after patching, since a fixed file on disk does nothing for a running process; if a restart must wait, disable the routine with -XX:-UseBASE64Intrinsics and verify via -XX:+PrintFlagsFinal.

more in playbook: Running AI in production
by read5 min views1 publishedNov 18, 2026
Oracle and OpenJDK patch Java4Shell, a critical Java flaw found with AI
Image: Startupfortune (auto-discovered)

If your servers run Java on recent Intel or AMD chips, a bug that lets a stranger run code on them without a password now has a fix. Whether you're exposed depends on a CPU detail most teams have never checked.

Oracle and the OpenJDK project shipped fixes on Tuesday, November 17, for CVE-2026-114938, a memory corruption flaw nicknamed Java4Shell. It sits in the code the Java Virtual Machine uses to decode Base64 quickly on x86-64 processors. According to Help Net Security, Oracle rates it 8.1 on the CVSS 3.1 scale because exploitation is hard, while Calif, the US security startup that found it, scores it 9.2 and calls it critical.

The fixed builds are OpenJDK 17.0.21.1, 21.0.13.1, 25.0.5.1 and 27.0.1.1, plus Java SE 8u513 for Oracle's performance builds, according to the advisory from CERT-EU. Affected distributions include Amazon Corretto, Eclipse Temurin, Microsoft Build of OpenJDK, Azul Zulu, SAP SapMachine, BellSoft Liberica and Red Hat OpenJDK. Check which of those your vendor ships, because the version numbers differ.

Fewer machines than the headlines suggest, but probably some of yours. CERT-EU says an attack needs an x86-64 CPU with AVX-512 extensions, which means Intel Xeon from Ice Lake onward or AMD Zen 4 and later, and attacker-controlled Base64 data reaching the decoder. That data can arrive through logins, JWTs, SAML assertions, cookies or TLS. ARM64 machines, including AWS Graviton instances, take a different code path and aren't affected. Neither are older x86 chips without those instructions.

Calif says it demonstrated unauthenticated code execution against Keycloak, Atlassian Jira and Confluence, JFrog Artifactory, Jenkins and PingFederate, and authenticated code execution against Apache Kafka, Solr, Elasticsearch, Gerrit and Spinnaker, as relayed by Help Net Security. Crashing a JVM is trivial. Turning that into a reliable takeover takes heap grooming and a memory leak, and VulnCheck and watchTowr both expect narrow, targeted attacks rather than mass scanning. No successful attacks have been reported yet.

Calif showed how AI can compress Apple exploit work into days Calif Global says its engineers used Anthropic's restricted Claude Mythos Preview to help build a macOS kernel exploit on Apple M5 hardware in five days. The story points to a bigger shift in offensive security, where small expert teams can use frontier models to compress work that once required far larger operations. - how to build macOS kernel exploits with AI - AI security research accelerating vulnerability discovery timelines

Calif's own description is blunter. Its researchers said that with careful exploitation the corruption can reveal sensitive memory, as Heartbleed did, or lead to remote code execution. Euronews, relaying the BBC's report, says Calif briefed Five Eyes partners and major cloud providers before going public.

What to patch and what to do if you can't #

Update, then restart every JVM. A patched file on disk does nothing for a process that's still running the old runtime. CERT-EU advises starting with internet-facing applications: identity providers, CI servers and wiki or ticketing tools are the obvious candidates given the list above.

If a restart has to wait, CERT-EU offers two stopgaps. You can turn off the vulnerable routine with the flags -XX:+UnlockDiagnosticVMOptions -XX:-UseBASE64Intrinsics, or you can cap vector instructions with -XX:UseAVX=2, which may cost some performance. You can confirm the first setting took effect with java -XX:+PrintFlagsFinal -version and searching the output for UseBASE64Intrinsics. CISA's advice adds input-size limits on Base64 data and watching for unexplained JVM crashes, which would be the first visible sign of someone probing. Security researcher Kevin Beaumont wasn't alarmed. His advice, quoted by Help Net Security, was to "just patch as you normally do later on today", adding that you should have done so last month because a different unauthenticated remote code execution bug was fixed then. If you skipped that round, this one lands on top of it.

The AI angle is real, but the outages aren't proven to be linked #

The BBC reports that Calif used AI to find the flaw, and that the same firm was involved in the PHP vulnerabilities patched in October. That's two widely deployed runtimes in a month, found by a small team using models rather than a big research lab. Expect more of it. Finding bugs just got cheaper, and every maintainer is going to be handed a longer queue.

It's tempting to tie this to this week's outages at banks, Apple and the FAA that some have blamed on rushed AI-assisted fixes. Nothing in the Java4Shell reporting supports that link. The fix here came through the normal coordinated process, with Oracle and OpenJDK publishing on the same day as the disclosure. The more useful lesson is about your own pipeline: if disclosures arrive faster, your window for testing a patch before you ship it shrinks, and teams that can't roll back a JVM update in minutes will feel it first.

Oracle hasn't said whether it has seen the flaw used in the wild.

Researchers find 1,465 exposed Artex AI hacking consoles open on the web Censys found 1,465 Artex consoles exposed online as of November 4, with 32 open directories leaking Windows lateral-movement tools, a sudo exploit and Linux implants that anyone could grab and reuse. - exposed Artex AI hacking consoles found on internet - Chinese AI penetration testing tool security breach discovery

Also read: Take-Two shares jump 7.5% as GTA VI hits 99 on Metacritic a day before launch • Palmer Luckey reveals SLAB, a seastead he says he has funded since selling Oculus • Tencent's third-quarter sales miss estimates as AI spending outruns forecasts

This article is posted in Technology News, check it out for more related stories.

Join the discussion #

Open in the community → Almost there. Sign in and your reply posts straight away.

── more in #ai-safety 4 stories · sorted by recency
── more on @oracle 3 stories trending now
sponsored brought to you by zahid.host 4,200+ EU-deployed projects
reading about agents? ship yours in a single git push.

Run your AI side-project on zahid.host

EU-based hosting, git-push deploys, automatic HTTPS, no cold starts. Free tier with a custom domain — perfect for shipping the agent you just read about.

$git push zahid main
→ Live at https://your-agent.zahid.host ✓
Get free account → Pricing
from €0/mo · no card required
LIVE [news/oracle-and-openjdk-p…] indexed:0 read:5min 2026-11-18 · —