Since Tuesday night I've had one question more than any other. "Oracle shipped the patch, so why are we still exposed?"
Because most companies don't get their Java from Oracle. They get it from a Java vendor, a Linux distribution, or baked into a product they bought. The fix has to work its way down that chain before you can install it, and right now a lot of it is still on the way.
What we're dealing with
CVE-2026-114938, or Java4Shell (some people are calling it Java2Shell), is a heap overflow in the Base64 decoder that ships with Java itself. Not a library. Java. Feed it Base64 with the padding in the wrong spot and it writes past the end of its buffer. On newer versions, 64 "A"s and an "=" will do it.
Calif, who found it, say they turned that into remote code execution against several widely used tools and platforms, default configs, no login needed. That includes the systems companies use to sign people in and to build their software. They also describe a memory leak along the way that can expose tokens and keys before any code runs. The trigger is public. The full exploits aren't, yet.
Oracle scores it 8.1 and says it's hard to exploit. Calif say 9.2. CISA told everyone to patch immediately. No one has reported attacks so far and it's not in CISA's KEV catalog.
You're affected if you run OpenJDK 17.0.6 or later, or anything on 21, 25 or 27, on an x86 server with a newer Intel or AMD chip, Ice Lake or Zen 4 onward. That's most x86 servers bought in the last few years. ARM is fine. So are Java 8 and 11, apart from one Oracle performance build.
The relay race
OpenJDK fixes the code. The Java vendors build their releases. The Linux distributions package them. If your Java lives inside a product, that vendor ships a new version with the fixed Java in it. Then it gets to you. Every handoff takes time.
Here's where things stood at 9am EST on Wednesday, November 18:
| Who | Status |
|---|---|
| Oracle Java SE | Fixed: 17.0.21.1, 21.0.13.1, 25.0.5.1, 27.0.1.1, 8u513-perf |
| Amazon Corretto | Fixed: 17.0.21.1.2, 21.0.13.1.2, 25.0.5.1.2 |
| Azul Zulu | Fixed: 17.0.21.1, 21.0.13.1, 25.0.5.1 |
| Eclipse Temurin | Not yet |
| Microsoft Build of OpenJDK | Not yet |
| Red Hat (and Rocky, AlmaLinux) | Rated critical, fix being expedited, not yet shipped |
| Debian | Not yet, all releases marked vulnerable |
| Ubuntu | Not yet, under evaluation |
The last leg is the slowest. Most of the tools and platforms Calif named hadn't put out an advisory by 9am EST Wednesday, and at least one still ships official container images with the vulnerable Java inside. I'm not knocking them. They can't start until the leg before them finishes, and then they have to build and test. But the patch you read about on Tuesday might be a while away from your servers.
What I'd do while you wait
Start by finding all of your Java. The package manager will show you the JDKs your team installed. It won't show you the ones inside products, appliances and container images, and those are the ones that tend to sit unpatched the longest, because nobody really owns them.
Then ask your vendors when their fixed build is coming. If they can't give you a date, you've learned something.
And turn on the workaround. One JVM setting switches off the vulnerable code:
Bash-XX:-UseBASE64Intrinsics
Add it to your JVM options or JAVA_TOOL_OPTIONS and restart. Base64 gets a little slower. I doubt you'll notice. You'll also see -XX:UseAVX=2 recommended, which turns off AVX-512 entirely. It works, but it costs more performance, and Java won't start with it on ARM, so think twice before you push it across a mixed fleet. Some providers are already doing this for you. At least one major cloud provider has applied it to some of its managed Java runtimes.
None of this replaces the patch. It covers you until the patch shows up.
What to fix first
I wouldn't put much weight on "hard to exploit." It's true, Calif say each target takes custom work. But think about what's in the line of fire. If someone gets into the system that signs everyone in, they don't have one server. They have everything that trusts it. Your build pipeline has your deployment credentials. Your artifact store has the software you ship to customers.
So my order would be: internet-facing login and single sign-on first, build pipelines and artifact repositories next, then the rest of your internet-facing Java, then everything else on 17 or later. And keep going until you're through the list.
Restart, then check
A new JDK on disk does nothing for a JVM that's already running. The old code stays in memory until the process restarts, and plenty of Java services run for weeks without one. So after you patch, or after you set the workaround, look at what's actually running:
Bashjcmd -ljcmd <pid> VM.versionjcmd <pid> VM.flags -all | grep UseBASE64Intrinsics
If you want to know whether a Linux box has the chip features the bug needs, check for all four of these:
Bashfor f in avx512f avx512bw avx512vbmi bmi2; dogrep -qw "$f" /proc/cpuinfo && echo "$f: yes" || echo "$f: no"done
Use that to decide where to start, not what to skip. In the cloud, the hardware under a workload can change without anyone on your team touching it.
Is it the next Log4Shell?
No. Log4Shell was one string in a log message and almost everything fell over. This needs the right Java, the right chip and a custom exploit for each app. That's a real bar.
But the trigger's public and so is the list of affected products. Our State of Vulnerabilities research found the median time from disclosure to exploitation is about five days now. With a lot of companies still waiting on the relay, I wouldn't bet on it staying quiet.
Where Mondoo fits
Mondoo finds Java runtimes across your servers, containers and cloud workloads, including the ones built into other products, and flags the ones affected by CVE-2026-114938. We rank them with our risk dimensions, so an internet-facing login system comes ahead of an internal test box on the same build. The details are on our CVE-2026-114938 page.
![]()


