Science & TechnologyLOT 261120 · 3 min read

Java4Shell flaw scores 9.2 on CVSS 4.0, exposes millions of services

Calif Research rates the OpenJDK Base64 flaw Java4Shell at 9.2 on CVSS 4.0. A 64-character input plus '=' triggers unauthenticated RCE on Ice Lake and Zen 4 servers.

By Sophie Lindqvist621 words

Key takeaways

  1. Java4Shell rates 9.2 on CVSS 4.0 per Calif Research; vendor scores range from 8.0 to 9.8.
  2. The trigger is 64 'A' characters plus '=' exploiting a bounds-check failure in OpenJDK Base64 assembly code.
  3. Only x86-64 CPUs with AVX-512 and VBMI are affected: Intel Ice Lake (2021) and AMD Zen 4 (2022) onward.
  4. Keycloak, Jira, Confluence, Jenkins and PingFederate have confirmed unauthenticated RCE.
  5. An AI bot found the flaw for roughly $200 in tokens, according to Calif Research.
Critical Java vulnerability puts millions of sites, services, and devices at risk of remote code execution
Critical Java vulnerability puts millions of sites, services, and devices at risk of remote code execution — AI-generated

A critical Java vulnerability rated 9.2 on the CVSS 4.0 scale by the lab that found it exposes millions of sites, services and devices to remote code execution. Calif Research, which discovered and dubbed the flaw Java4Shell, says the bug sits in OpenJDK — the library from which nearly every Java distribution derives its implementation, including Oracle's own.

The practical consequence is severe. An attacker can often gain code execution without logging in to the target application, then read user and internal data and use the system as a launch point for further access, including database connections and potential full system takeover. Calif Research's 9.2 rating is not universal, however. Oracle and Red Hat each issued their own scores, and vendor assessments range from 8.0 to 9.8 depending on who is counting. That spread alone deserves scrutiny.

What triggers the bug?

The root cause is a flaw in a hand-written assembly optimization for Base64 decoding, a function so ubiquitous that almost no modern application avoids it. Base64 transforms binary data into plain text. It embeds image data inside HTML and CSS, attaches files to email messages, and represents checksums and cryptographic key fingerprints as readable text.

The trigger is almost comically simple. An attacker inputs 64 "A" characters followed by a single "=" sign as padding. The faulty optimization fails a simple bounds check, overwrites neighboring memory, and lets the attacker weaponize the corruption. Calif Research compares the mechanism to Heartbleed, the 2014 OpenSSL flaw that leaked memory contents across the internet.

There is a hardware condition. The vulnerable Java instance must run on an x86-64 CPU with AVX-512 support and the VBMI extension. In practice, that means Intel Ice Lake server chips from 2021 and AMD Zen 4 server chips from 2022 onward. Older CPUs are not affected, because the buggy optimization is specific to VBMI. That constraint narrows the blast radius — but only somewhat, since the newest servers are precisely where the most valuable workloads run.

Which products are exposed?

Calif Research listed the confirmed casualties at the time of its disclosure:

  • Full unauthenticated RCE: Keycloak, Jira, Confluence, Jenkins, PingFederate
  • RCE after authentication: Apache Kafka Broker, Solr, Elasticsearch, Gerrit, Spinnaker

The list is unlikely to be complete. Operating systems, digital certificates and individual programs frequently bundle their own copy of a Java executable, so the real exposure extends far beyond dedicated Java deployments. IBM, Red Hat, Siemens and Broadcom have all had to issue security advisories and updates.

Containers compound the problem. A large share of container images across public registries include Java-based applications in some form, and Calif Research expects the count of vulnerable images to be high. Oracle has already patched the bug and released updated versions of its software, but patch propagation through embedded copies and container registries will take far longer than the fix itself. This is unlikely to be a bug that gets eradicated quickly.

How was it found?

Calif Research says an AI bot found the vulnerability, at a cost of roughly $200 in tokens. Unlike many AI-assisted security claims, this discovery was not low-hanging fruit, according to the researchers — the bug involved genuinely deep analysis rather than evidence of poor code or network hygiene. The claim is notable, and independently verifying how much of the find was machine work remains difficult.

For infrastructure teams, the immediate arithmetic is straightforward. Anything running a modern Java stack on Ice Lake or Zen 4 hardware, exposed to user input, should be treated as compromised until patched. Oracle's fixes exist today; the industry's installation of them will define how long Java4Shell remains a live threat, and history suggests that tail runs for years.

Tom's Hardware (Source)

Filed under

Share this article:

More from Sophie Lindqvist

Sophie Lindqvist

Show full bio

News editor covering money, funding and market shifts around Semiconductors, chip manufacturing, hardware, AI compute at Chip Dispatch.

1,188 articles

Read also

« Previous article