DevelopingJava4Shell (CVE-2026-114938): an out-of-bounds write in the Java runtime's Base64 decoder, patched Monday, no exploitation seen yet. What to patch, in what order, and how to prove it
The bug is in the JVM, not a library, and it runs before your code gets a vote; that makes it an inventory problem first and a patching problem second.

On Monday Oracle's November Critical Patch Update shipped fixed builds of Java SE 8, 11, 17, 21, 25 and 27 and of GraalVM, and the firm that reported the underlying bug, Calif, gave it a name: Java4Shell. The CVE record for CVE-2026-114938 had not been published when the patches went out, which is why the numbers in circulation disagree. Early reports score it 9.2 under CVSS 4.0; Oracle, which scores under CVSS 3.1, lists nothing above 8.1 for Java SE in this update, and Aikido has pointed out the two figures may describe different issues. What is agreed is the shape of the flaw: an out-of-bounds write in OpenJDK itself, reached through the runtime's hardware-accelerated Base64 decoding, which processes bytes before an application has any chance to validate them. An unauthenticated attacker who can get malformed Base64 in front of a vulnerable JVM may be able to run code.
Who is exposed
The accelerated routine only runs on x86-64 processors with AVX-512 VBMI, so ARM fleets and older Intel and AMD parts are outside the bug's reach even if they run a vulnerable build. Inside that boundary the scope is wide. Early reports name OpenJDK 17.0.6 and later, and Oracle's update covers Java SE 8u511, 11.0.33, 17.0.21, 21.0.13, 25.0.5 and 27.0.1 plus GraalVM for JDK 17 and 21 and GraalVM Enterprise 21.3.20. Distribution packages, Eclipse Temurin, Amazon Corretto, Azul Zulu, Microsoft and Red Hat builds, and the JDKs quietly bundled inside appliances and container images all inherit the flaw, under version numbers that will not match Oracle's. Datadog Security Labs characterises it as a runtime flaw rather than a library one, which is the detail that should set your response: there is no dependency to bump, and a software bill of materials that lists applications rather than runtimes will not surface it.
What is not yet known
The vulnerable component and the exact exploitation conditions were not public as of Monday, and Calif says it will publish details after patches have had time to land. Cloudflare reports no signs of exploitation across its network and Aikido had seen no reports by Monday evening. That window is the useful part of this story. Log4Shell went from disclosure to mass scanning in hours because the trigger string was trivial; here the trigger is a malformed Base64 payload whose precise form is being withheld, so defenders have days rather than hours, but not more.
What to do
First, find the JVMs. Count running processes, not installed packages: jcmd <pid> VM.version on a host tells you what is actually executing, and container images need the same check inside the image. Second, patch in exposure order. Oracle's fixed builds are 8u513, 11.0.33.1, 17.0.21.1, 21.0.13.1, 25.0.5.1 and 27.0.1.1; other vendors will ship their own numbers. Anything that accepts Base64 from the internet goes first, which in practice means API gateways validating JWTs, services handling WebAuthn, basic-auth headers, data-URI uploads and multipart bodies. Third, restart. Upgrading a package does not replace a running JVM, and a fleet that reports 'patched' from package inventories while the old process keeps serving traffic is the failure mode to expect. Fourth, use the compensating controls while you work. Cloudflare has published six Java4Shell rules in its managed ruleset, covering malformed Base64 in headers, query strings, bodies, path segments, base64url JWT components and WebAuthn values, all defaulting to log-only on paid plans; it is also rolling out a JVM start-up option that disables the accelerated routine and falls back to the standard decoder, which it began deploying about an hour after the patches appeared. Cloudflare calls these a compensating control, not a fix, and that is the right framing: they narrow the path, they do not close it.
What to watch
The CVE record, when it is published, will settle the score and the affected-version list. Calif's technical write-up will mark the point at which exploit development becomes easy; Project Zero's new disclosure clock is a reminder that these windows are shrinking. And watch your own telemetry for Base64 decoding errors spiking on edge services, which is the cheapest early signal that someone is probing.
The story so far
- Nov 16, 2026 ARTEX inside AWS: anonymous S3 reads, CloudWatch log hunting with stolen keys and self-registration on open Cognito pool…
- Nov 15, 2026 FBI FLASH on ARTEX: under three hours to privileged access, 67,000 South Korean records, 2,250 ConnectWise tenants and 1…
- Nov 13, 2026 AntiFox: attackers ran an open-source AI pentest platform from Alibaba Cloud to steal AWS secrets, hundreds of API calls…
- Nov 12, 2026 Railway's post-mortem: a build-sandbox escape, one over-trusted credential and $740,000 of fraudulent AI charges
Key terms in this story
- Out-of-bounds write A memory-safety bug in which a program writes data past the end of the buffer it was given, letting an attacker corrupt memory and often run code.
- Full X disclosure (Project Zero) Google Project Zero's 2026 policy: a vulnerability it reports is published X days later whether or not a fix exists; X is 90 now, 60 from February 2027 and 30 from January 2028.
- AVX-512 A family of Intel and AMD processor instructions that operate on 512 bits at once, used to speed up tasks such as encoding, encryption and machine-learning maths.
More in Security
All →
Microsoft's agent-framework package was poisoned on PyPI for 45 minutes with a stealer that ran at every Python start-up. Two versions, one webhook, at most 50 victims, and a checklist for this morning
The attack did not need you to import the package. A .pth file in site-packages ran the stealer whenever any Python interpreter launched, which is why…

Google's GTIG finds 'exploit_agent', the first framework that turns a vendor patch into a live exploit with no human in the loop. Zero-days hit 22 a month in August. What to change in your patch and detection plan
Google's fourth-quarter AI threat report moves the attacker's clock to the moment a patch is published, documents a backdoor delivered through a codin…

Project Zero's 'Full X' policy: 90 days now, 60 from February, 30 from January 2028, fixed or not. What LLM-speed exploitation means for your patch SLAs
Google's bug-hunting team has scrapped the 30-day soak period it added in 2021 and will publish vulnerability details on a fixed clock because, it say…
Get the brief in your inbox
One email every morning: the stories that mattered in AI, each with a one-line "why it matters" for your role. Free, no ads, unsubscribe any time.
Published daily by AI Briefing. We never share your address.