Firevault, offline secure storage, physically disconnected from the internet
Pricing
A dark data centre aisle with one server rack showing a magenta warning light beside a monitor displaying blurred code with one highlighted line

BreakingNews

Java4Shell: critical Java flaw found with AI puts enterprise applications at risk

A critical flaw in Java, found by Calif using AI, can let attackers read memory or run code on applications such as Jira, Jenkins and Keycloak. Oracle has released fixes and CISA urges immediate patching. What is affected and what to do now.

Mark Fermor, CTO, CMO and Founder of Firevault

Mark FermorCTO, CMO and Founder, Firevault

Published 18 November 2026 · 5 min read

What happened

A critical security flaw has been found in Java, the programming language and runtime behind a large share of the world's business software, the BBC reported on 18 November 2026.

The flaw, named Java4Shell and tracked as CVE-2026-114938, was found by the security start-up Calif, the same firm that identified the PHP2Shell flaw in October. Calif says it used AI to find and develop the attack, and the BBC reports that the model involved was OpenAI's GPT-6 Cyber, which is not yet publicly available.

Oracle, which owns Java, released fixed builds as part of its critical patch update on 17 November. The US Cybersecurity and Infrastructure Security Agency (CISA) has urged organisations to patch immediately.

What the flaw does

Java includes a shortcut for decoding Base64, a common way of turning files, login details and network messages into text. On certain modern processors, Calif found that malformed input can make that shortcut write past the end of its memory before Java rejects the input. With careful work, an attacker can use this to read sensitive memory, which Calif calls "JavaBleed", or to run their own code on the server.

Calif says its testing achieved code execution without logging in on Keycloak, Atlassian Jira and Confluence, JFrog Artifactory, Jenkins and PingFederate in their default settings, and after login on Apache Kafka, Solr, Elasticsearch, Gerrit and Spinnaker. It says the list is not exhaustive. Calif scores the flaw 9.2 out of 10 under CVSS 4.0; Oracle scores it 8.1 under CVSS 3.1.

Exposure depends on both the software and the hardware. According to Calif and Tenable, only affected Java versions (OpenJDK 17.0.6 and later, including Oracle Java, Amazon Corretto, Eclipse Temurin, Microsoft, Azul and Red Hat builds) running on x86-64 processors that support AVX-512 VBMI are affected. ARM processors are not.

Is it being exploited?

As of 18 November, Tenable reports that GreyNoise has seen exploitation attempts but has not confirmed any successful attack against its sensors. VulnCheck and a researcher at Immersive say they have built working exploits. VulnCheck told SecurityWeek that crashing an affected application is easy, but that running code needs substantial work tailored to each target, and that it does not expect the flaw to be "easily or widely exploitable".

The BBC reports that it is unclear whether any platforms or companies have been affected. The BBC also notes that the National Security Agency did not say whether its recent network outage was linked to the bug, and that the US Secretary of State said there was "zero evidence" that this week's US technology incidents were related.

Professor Toby Walsh told the BBC the flaw had been "hiding in plain sight for 12 years". Calif's own list of affected versions begins with OpenJDK 17.0.6, released in 2023, so exactly how long this specific weakness has existed is not yet clear.

A pattern, not a one-off

Two critical flaws in two of the world's most widely used languages, found within a month by the same firm using AI, tell us something important. Walsh told the BBC that "AI is now exposing more flaws much more quickly", including ones humans might not have found alone.

The same tools are available to attackers. The time between a flaw becoming known and someone trying to use it is shrinking, and for Java4Shell, Calif says news of the flaw leaked before its planned disclosure.

Patching remains essential. But every organisation should now plan on the assumption that the software it depends on will, at some point, contain a flaw nobody has found yet.

What to do now

  1. Find your Java. CISA advises reviewing software inventories for Java runtimes, including those bundled inside third-party applications, container images and appliances, which may not appear as separate entries.
  2. Patch, starting with internet-facing systems. Apply Oracle's fixed builds (17.0.21.1, 21.0.13.1, 25.0.5.1 and 27.0.1.1) or your distribution's equivalent, and vendor updates for the applications listed above.
  3. If you cannot patch yet, use the interim measures. Calif identifies the Java startup option -XX:-UseBASE64Intrinsics, which Keycloak recommends as an interim mitigation. CISA also advises limiting unauthenticated access and enforcing input-size limits.
  4. Watch for warning signs. CISA lists unexpected Java crashes, rogue runtimes and suspicious Java processes as indicators to monitor.
  5. Rotate secrets on exposed systems. Because the flaw can leak memory, treat keys and tokens held by affected, internet-facing applications as potentially exposed.

Where Firevault fits

No storage or access control product removes a flaw inside Java. The fix is the patch.

What Firevault addresses is what a compromised system can reach next. Jira, Confluence, Jenkins, Artifactory and Keycloak sit at the centre of how organisations build software and manage identity. A foothold in one of them can lead to source code, credentials, backups and the systems they connect to.

Offline Secure Storage® keeps selected critical data and recovery copies on dedicated hardware that is physically disconnected and offline by default. If data is connected, it can be reached. Data held offline is not reachable through a compromised Java application, so a clean copy remains available if connected systems and their backups are affected.

Control by Firevault applies just-in-time access to the connections that suppliers, administrators and integrations hold into an environment, so those paths exist when approved work needs them and are physically disconnected the rest of the time. Our Plans for controlling third-party access and interrupting data exfiltration explain the approach.

Patch what is vulnerable. Disconnect what does not need to be connected. Protect what matters.

Mark Fermor is CTO, CMO and Founder of Firevault. Learn why offline storage is different or talk to the team.

Questions

Java4Shell, tracked as CVE-2026-114938, is a memory-corruption flaw in the way Java decodes Base64 data on certain x86-64 processors. Calif, which found it, says an attacker can use it to read sensitive memory or run their own code on affected servers, in some applications without logging in.

According to Calif and Tenable, affected Java versions are OpenJDK 17.0.6 and later, including builds from Oracle, Amazon, Eclipse, Microsoft, Azul and Red Hat, running on x86-64 processors that support AVX-512 VBMI. ARM processors are not affected. Calif demonstrated attacks on applications including Jira, Confluence, Jenkins, Keycloak and JFrog Artifactory.

As of 18 November 2026, Tenable reports that GreyNoise has seen exploitation attempts but has not confirmed a successful attack against its sensors. Working exploits have been built by security firms, and VulnCheck says running code requires substantial work for each target. The BBC reports it is unclear whether any organisations have been affected.

Calif says it used AI to find and develop the attack. The BBC reports that the model involved was OpenAI's GPT-6 Cyber, which is not yet publicly available. Calif also found the PHP2Shell flaw in October 2026.

Identify every Java runtime, including those bundled in third-party applications and containers, apply Oracle's or your distribution's fixed builds starting with internet-facing systems, and update affected applications. Where patching must wait, Calif identifies the startup option -XX:-UseBASE64Intrinsics as an interim measure. Rotate keys and tokens held by exposed systems.

No. The fix for a flaw inside Java is the patch. Offline Secure Storage keeps selected critical data and recovery copies physically disconnected, so they are not reachable through a compromised application, and Control by Firevault keeps supplier and administrator connections disconnected until approved work needs them.

Mark Fermor, CTO, CMO and Founder of Firevault

About the author

Mark Fermor

CTO, CMO and Founder of Firevault. Mark writes on physical control, offline storage and the incidents that show why both matter.

Offline Secure Storage icon

Why #OSS

Keep a clean copy out of reach.

Offline Secure Storage® keeps one copy physically disconnected at the hardware level between sessions.

Control by Firevault icon

Control

Govern the paths attackers depend on.

Physical, out-of-band control of selected network paths, composed into Control Plans and industry Blueprints.

Vault

Ready to start?

Create your Vault online.

Business Vaults from £360 a month inc VAT, on a 36-month commitment.