A serious flaw in Java, found with the help of an AI model, has landed in the middle of a run of technology outages and a growing argument about AI-found vulnerabilities. The bug, nicknamed Java4Shell, was disclosed on November 17 and prompted an urgent patching alert from the U.S. Cybersecurity and Infrastructure Security Agency (CISA). Whether AI-driven patching is behind the recent outages at banks and the National Security Agency is disputed, and officials have said the incidents were not a coordinated cyberattack.
No single report covers all of this. This article draws on CISA's alert, reporting by Fortune, Engadget and The Next Web on the Java flaw, and an earlier Reuters report on AI and software vulnerabilities. Where a claim rests on one outlet, we say so.
What Java4Shell is
CISA's alert, released November 17, identifies the flaw as CVE-2026-114938, a memory-corruption bug in the Base64 decoder of the x86-64 HotSpot Java Virtual Machine. In CISA's words, specially crafted Base64 input could crash the JVM and allow unauthenticated remote code execution. The agency strongly urged organizations to patch immediately. The Canadian Centre for Cyber Security contributed to the alert.
Java runs inside a large share of enterprise software, which is why the alert matters. CISA told organizations to inventory every JDK and Java runtime, including those bundled with third-party applications, container images and appliances, because Java may not appear as its own line in a software inventory. It said to prioritize internet-facing systems.
Severity is rated differently depending on who is scoring. The Next Web reported that Oracle rates the flaw 8.1 out of 10 and hard to exploit, while the security startup that found it, Calif, rates it 9.2. Oracle released a critical patch update on Tuesday, although Engadget noted the update did not specifically name this bug.
How AI found the Java flaw
Engadget reported that Calif found the flaw with help from OpenAI's GPT-6 Cyber model, which is not publicly available. Feeding the decoder a string of 64 "A" characters followed by an equals sign triggers a memory error, which usually crashes the program but on some machines can be exploited to gain unauthorized access. Calif showed the bug could be used to break into Keycloak, a login system used by organizations including CERN and NASA.
The Next Web, a smaller outlet, added detail that we could not confirm elsewhere. It said a Calif researcher used OpenAI's Codex agent to find the flaw in about an hour, then built a first working exploit within two more hours for under $200 in AI tokens, and that The New York Times independently verified the attack. The same report says the flaw affects OpenJDK 17.0.6 and later on Intel and AMD chips with AVX-512 VBMI, and does not affect ARM chips.
Calif says the bug is not easy to weaponize, and there is no universal exploit, so each application needs its own attack, according to The Next Web. The firm is withholding product-specific exploits until users can upgrade. Calif estimates the bug could linger for five to 10 years because some systems will go unpatched, Engadget reported.
The disclosure drew criticism. The Next Web reported that Jan Schaumann of Akamai accused Calif of "publicity-maxxing" on the Openwall mailing list, and that the trigger input had been posted in a proposed OpenJDK fix in September before it was closed without merging.
The outages and what is actually linked
Fortune reported on November 18 that the NSA suffered an internal communications outage and that Bank of America, Wells Fargo and JPMorgan Chase all had outages the same week. A spokesperson for Sen. Mark Warner linked the NSA and banking outages to botched patch implementation. Officials said the incidents were not a coordinated cyberattack.
The explanation is contested. Fortune attributed to The Washington Sun a report that bank CEOs and the NSA told Congress that AI-driven patches were to blame, and Fortune did not independently confirm it. A JPMorgan Chase spokesperson said OpenAI's tools were not a factor and that its problem was "a temporary issue related to a software bug in on-premise hardware systems," not something found or fixed with AI. The Next Web likewise noted it could not confirm the cause of the outages.
Palo Alto Networks CEO Nikesh Arora wrote on X that "Some of the recent outages are a consequence of a deluge of patches." Cliff Steinhauer of the National Cybersecurity Alliance said teams "don't always have time to fully test every update." Amit Shah, CEO of InstaLILY AI, called it a failure of implementation. None of these statements shows that the Java flaw caused an outage, and the reporting we reviewed does not claim it did.
Why AI-found vulnerabilities are piling up
A Reuters report published on October 23 set out the underlying trend. Seven cybersecurity professionals told Reuters they expect AI-driven discoveries to cause more outages, and VulnCheck said AI had helped double the number of software flaws found year on year. The report tied a wave of European government website outages to weaknesses found in PHP, with some operators taking sites offline because they could not patch before attackers struck.
Shane Fry, chief technology officer of RunSafe Security, said what is changing is the timescale. Matthew Prince, CEO of Cloudflare, predicted "a couple of years where every single week it's going to feel like the sky is falling," while saying the cleanup would be a net positive. Experts quoted by Reuters disagreed on whether the flood of disclosures would lead to widespread hacking. Ramy Rahman of ArmorCode said more disclosed vulnerabilities do not automatically mean more successful attacks.
What organizations are doing about it
Fortune cited a CNBC report from October 14 that Bank of America hired 1,000 contractors to patch flaws exposed by Anthropic's Mythos model, on top of 3,000 internal staff. We did not review the CNBC article directly. It shows the scale of effort at one large bank.
For Java specifically, CISA recommends applying Java and vendor patches as soon as they are available. If patching cannot happen immediately, it suggests validating and restricting Base64 input at the application boundary, enforcing input-size limits and limiting unauthenticated access. It also advises monitoring for rogue runtimes, unexpected JVM crashes and suspicious Java processes.
What remains unclear
Several questions are open. The Next Web said the security firm watchTowr reported seeing exploitation attempts, while other labs had reproduced crashes and memory leaks but not code execution. Whether attackers are using the flaw at scale has not been established in the sources we reviewed. The cause of each outage is also unsettled, and the evidence so far points to patching and implementation problems more than to attacks.
The broader tension is speed. AI tools can find and exploit flaws faster than many teams can test and deploy fixes, a point Jeff Collins of WanAware made to Fortune. How companies balance fast patching against stable systems will decide whether more outages follow.
Frequently asked questions
What is Java4Shell?
It is the nickname for CVE-2026-114938, a memory-corruption flaw in the Base64 decoder of the x86-64 HotSpot Java Virtual Machine. CISA says crafted input could allow unauthenticated remote code execution.
Who found it and how?
The security startup Calif found it using OpenAI's GPT-6 Cyber model, according to Engadget. The Next Web reports a researcher also used OpenAI's Codex agent.
Did the Java bug cause the bank and NSA outages?
The reporting we reviewed does not say so. Outages were linked by some officials and executives to rushed patching, but JPMorgan said AI tools were not a factor in its outage, and the causes remain disputed.
What should organizations do now?
CISA advises inventorying all Java runtimes, applying patches promptly and prioritizing internet-facing systems. If patching must wait, it suggests restricting Base64 input and limiting unauthenticated access.
Will AI cause more software vulnerabilities to be disclosed?
VulnCheck told Reuters that AI helped double the number of flaws found year on year, and experts expect more disclosures. They disagree on whether more disclosures will lead to more successful attacks.



