Back Bankinfosecurity AI Agent Finds Critical Java Flaw in Just 3 Hours
Artificial Intelligence & Machine Learning , Governance & Risk Management , -Generation Technologies & Secure Development
Just weeks after the discovery of PHP2Shell, artificial intelligence models found another critical vulnerability, this time in Java.
See Also: Introducing Unit 42 Frontier AI Defense: A Defender's Guide to the Frontier AI Impact
Cybersecurity startup Calif said in a blog post Tuesday that it discovered CVE-2026-114938 , a memory-corruption vulnerability in OpenJDK's Base64 decoder, a popular Java runtime platform. It calls the flaw Java4Shell.
Since Calif's disclosure, Oracle said it released a patch for the vulnerability.
Java is among the most popular programming languages on the internet, powering more than 90% of the world's largest companies, including many government systems.
Calif said the vulnerability can reveal sensitive memory and allow for remote code execution. It added that it can trigger the corruption "by just 64 A characters followed by a single =."
In discovering the flaw, Calif researchers only spent $200 in tokens, and it took three hours - one to discover the vulnerability and two to create an exploit.
"We discovered a bug in that optimized machine-level code that lets malformed input overwrite neighboring memory," the company said.
Calif found the vulnerability after one researcher, Quang Luoung, audited OpenJDK and chose to look at the platform's compiler, cryptography and APIs. Luoung directed Codex, OpenAI's coding agent, to look at each target "like any serious security researcher in 2026." The company did not disclose which OpenAI model ran on Codex during its investigation, but The New York Times reported it was the recently released GPT-6 Cyber.
While Luoung considered a large chunk of Codex's responses to be "slops," it found memory corruption in one of Base64's decoders.
Calif said one of the selling points of Java is its memory safety, but because Java is written using languages like C and C++, memory corruption persists. The company said the bug "sits in the hand-rolled assembly code that handles the final few characters" of the exploit.
Calif said there is no universal exploit for Java4Shell, noting that it depends on how the organization uses Base64.
"Calif was able to accelerate the exploit development by spending initial effort building pipelines and strategies, then directing AI agents to use those for initial exploitation. Afterward, most of the later chains are built almost autonomously with AI," the company said.
British cybersecurity expert Kevin Beaumont advised his social media followers to stay calm. "There's a bunch of TLP RED bonkers briefing a world ending Java vuln," he wrote . "It's not world ending. It's not very exciting. Just patch as you normally do later on today - as you should have last month as there was a different unauth RCE vuln then."
Java4Shell is the most recent critical vulnerability found around a language that many websites rely on. PHP2Shell, disclosed on Oct. 21, impacted the very popular PHP, which powers not just websites but the hosting platforms that support them.
AI model providers and cybersecurity companies have urged organizations to use AI agents to defend against threat actors, especially as AI models and agents increasingly reveal vulnerabilities and make it easier for hackers to attack systems.
The full story
This article is one source in a clustered incident — the cluster page carries the summary, timeline and every other outlet covering it.
