Application security, Third-party code, Security Operations, SOC, Vulnerability Management

Java4Shell flaw: Attackers probe critical JVM vulnerability

Colorful lines of code seen through a magnifying glass on a dark screen, symbolizing careful debugging, code review, and detailed software quality analysis

Even as security teams across practically every vertical on Nov. 18 were working hard to patch the Java4Shell flaw, security experts said it’s just a matter of time before the Java Virtual Machine (JVM) bug gets exploited.

“We are already seeing attackers spray the internet and hit our global honeypot network with requests decorated in 'java4shell-recon' User-Agents, alongside blind exploitation attempts using correctly shaped requests,” said Jake Knott, head of threat intelligence at watchTowr.

“What we’re seeing so far isn’t particularly functional or scary yet, but it shows the clear signal that attackers want to leverage this vulnerability," said Knott. "Over the coming days, we expect attackers to iterate product-by-product to refine and build more reliable target-specific exploits, and that's where we start to see real exploitation happen.”

The flaw — CVE-2026-114938 — was patched yesterday by Oracle as well as flagged by the Cybersecurity and Infrastructure Security Agency (CISA), which put out an advisory on Java4Shell late yesterday.

Keep in mind that it’s estimated that 98% of companies use the JVM for some part of their development efforts, so this latest flaw caused great concern given all the recent outage events surrounding airports, banks, email infrastructure, and yesterday’s report of an outage in NSANet, the National Security Agency’s classified internal communications network.  

As of this afternoon, there was still no known “in the wild” exploitation on Java4Shell, but security pros warned teams to patch right away.

“There’s no confirmed in-the-wild exploitation or public RCE exploit, only lab demonstrations,” said Seemant Sehgal, founder and CEO at BreachLock. “But the details leaked before the patch and researchers built a working exploit within hours, so treat this as a 'when,' not an 'if.'”

Sehgal described Java4Shell as a heap out-of-bounds write in the JVM’s AVX-512 Base64 decoder, where a short malformed string can corrupt memory and, with an application-specific chain, lead to unauthenticated remote code execution (RCE). Sehgal said it’s rated 9.2 under CVSS 4.0 and has been demonstrated in the lab against Keycloak, Jenkins and Jira, the identity and build systems everything else trusts.

“Oracle has shipped fixes and other OpenJDK vendors are following, and the price of patching is an update and a restart,” sid Sehgal. “The cost of waiting is betting your SSO and CI/CD pipeline on attackers being slower than the researchers were.”

Adrian Culley, offensive security engineer at SafeBreach, added that while as of now, he’s seen no credible report of a successful compromise using Java4Shell, that’s not the same as quiet.

“Sensors are reported to be picking up requests shaped like the trigger, a researcher outside the finding team says he has a working exploit for a specific application, and unverified exploit code is appearing on GitHub,” said Culley. “CISA’s alert is a patch-now alert, not a known-exploited listing.”

Culley explained Java4Shell as a memory-corruption bug in the JVM itself, in the hardware-accelerated Base64 decoder HotSpot uses on newer Intel and AMD server chips. Culley said Base64 is everywhere an attacker can reach before logging in: authorization headers, tokens, and certificates.

“So a memory-safe language inherits a classic heap overwrite, in identity providers, build servers and artifact stores,” said Culley. “Crashing an affected service is trivial. Reliable code execution needs a separate chain per application, which is why I’d expect targeted use against high-value systems rather than a worm. What should worry people is the researchers’ own account: about an hour to find with AI assistance, two more to a working exploit. Withholding the chains buys days, not months.”

Saeed Abbasi, head of the Qualys Threat Research Unit, said updating the Java runtime is the only complete fix.

If the team cannot patch today, Abbasi said there’s a temporary workaround: it’s a startup setting for Java, applied by the team that runs the application, and it takes effect after a restart. The team needs only one of these two options:

  • Preferred: -XX:-UseBASE64Intrinsics. This turns off the fast Base64 decoder where the bug lives. Java falls back to its standard decoder, and everything else runs as before. Applications that decode large amounts of Base64 may run slower, so test first.
  • Alternative: -XX:UseAVX=2. This turns off a wider set of processor optimizations, including the vulnerable one, so the performance cost is broader. Use it only on Intel and AMD servers. Java on ARM does not accept this setting and will not start.

“Neither option replaces the patch,” said Abbasi. “Either way, restart every Java process. A patch or a flag takes effect only when the JVM restarts, so verify the running process and not only the installed package. Do not rely on a reverse proxy as a control. Request buffering breaks some exploit chains but is not a general mitigation.”

Meanwhile, Abbasi said teams should reduce exposure while they wait for a patch: take identity, build, and collaboration consoles off the open internet, or put them behind a VPN, until they are patched. CISA also suggests validating Base64 input and enforcing input-size limits at the application boundary. Size limits alone will not stop this bug, said Abbasi, because the malformed input is very short.

Sina Chehreghani, head of threat intel and threat research at Securonix, added that Calif said in a blog it found the flaw with AI assistance while auditing OpenJDK, then built an initial exploit in about two more hours. That points to a faster research cycle: AI can help examine large codebases and investigate obscure execution paths, while researchers still need to validate the bug and show how an application can reach it.

“We should expect more findings in mature, widely used software, especially in optimized code that receives untrusted input,” said Chehreghani. “The pressure will fall on maintainers to verify reports and coordinate fixes across runtime vendors, and on organizations to know which Java versions, bundled runtimes and processor types their applications actually use.”

Steve Zurier
Steve Zurier has been a freelance writer and editor for SC Media since 2012. Now, Zurier writes daily news stories and edits SC Media’s Perspectives columns. A long-time member of the tech press, Zurier lives in Columbia, MD. During off-hours, Steve moonlights as an upright bassist for jazz and klezmer bands around the Baltimore/DC area.

Get daily email updates

SC Media's daily must-read of the most current and pressing daily news

By clicking the Subscribe button below, you agree to SC Media Terms of Use and Privacy Policy.

You can skip this ad in 5 seconds