Calif Unveils Java4Shell, Vulnerability Detected by AI in One Hour

Calif announces that its researcher Quang Luong discovered Java4Shell, a critical vulnerability in OpenJDK, in just one hour with the help of AI. Rated 9.2 out of 10 on the Common Vulnerability Scoring System version 4.0, this memory corruption can allow for data theft or code execution. However, its exploitation is subject to strict conditions regarding hardware, software, and applications, security researchers emphasize.
Oracle Releases Patches on November 17
Oracle and the OpenJDK vulnerability group released their patches in a coordinated manner on November 17. Calif had delayed the publication of its complete technical explanations until they were made available. CISA urges organizations to take immediate action, prioritizing systems accessible from the Internet.
The agency recommends inventorying JDKs and Java environments, including those embedded in third-party software, and installing Java patches from vendors as soon as they are available. It advises consulting Oracle's security advisory to understand the affected versions and products as well as the remediation procedures. Monitoring should focus on unwanted runtime environments, suspicious Java processes, and unexpected JVM crashes: according to CISA, a malicious entry can also cause service interruptions.
When an immediate fix is not possible, CISA recommends temporary protections: controlling and restricting Base64 data as soon as it enters the application, capping its size, and reducing unauthenticated access. The Swiss National Cyber Security Centre, NCSC, encourages operators to clarify their exposure with their suppliers. For individuals and end users, it recommends quickly installing security updates using the official functions of programs and devices. Depending on the product, fixing the embedded Java component involves updating the software or device.
Calif Demonstrates Attacks on Multiple Applications
Calif claims to have achieved remote code execution without prior authentication on Atlassian Jira, Confluence, Jenkins, JFrog Artifactory, and Keycloak. These tests involved unpatched installations retaining their default settings. Under the same conditions, the company also reports attacks requiring authentication on Apache Kafka Broker, Apache Solr, and Elasticsearch. NCSC warns that other Java applications could be affected, including banking software and tools specific to certain sectors.
For Justin Moore, Director of Adversarial Operations at Arctic Wolf, a successful exploitation could grant the attacker the rights of the compromised application. These rights could provide access to sensitive information, credentials, cloud resources, identity systems, or development infrastructures. When the software performs a critical business function, he believes the consequences could extend beyond the initially compromised system.
Le brief IA que les pros lisent chaque soir
Les 7 actus IA du jour, décryptées en 5 min. Gratuit.
Inclus dès l'inscription : notre sélection des meilleurs guides & comparatifs IA.
Choisissez votre rythme
Gratuit · Pas de spam · Désabonnement en 1 clic
Saeed Abbasi, head of the threat research unit at Qualys, points out that an attack can read the Java memory heap even before executing hostile code. Private keys, tokens, or personal data may be present, he notes. Chris Fistonich, Senior Director at Malwarebytes Threat Labs, considers the vulnerability somewhat more difficult for cybercriminals to monetize but believes that the respite is measured in days rather than months.
The Decoder Exposes Java Memory
Listed under the identifier CVE-2026-114938, Java4Shell affects the x86-64 Base64 decoder of OpenJDK, the most popular Java runtime environment. Specifically crafted data can breach memory boundaries. The flaw can reveal information from the memory heap, with a mechanism similar to Heartbleed, or allow arbitrary code execution through web requests and authentication exchanges.
Java was designed to prevent memory corruption. However, some components for which performance is critical still rely on low-level machine code. As the third most popular programming language globally, Java is used by 90% of Fortune 500 companies. It is present in banking, aviation, telecommunications, and national security systems.
AI Accelerates Vulnerability Research
After identifying the flaw during an AI-driven code audit, Quang Luong reportedly spent an additional two hours on the first functional exploit, according to Calif. The company estimates the total expenditure in tokens to be less than $200. It had already significantly contributed to the discovery of the PHP2Shell memory corruption vulnerabilities that emerged last month.
Google recently warned that attackers are now fully automating the search and exploitation of vulnerabilities using AI, without human intervention. Google and Anthropic announced they would shorten their coordinated disclosure timelines, with the traditional 90-day embargo potentially dropping to 30 days. Natalie Silvanovich, a security engineer at Google Project Zero, believes that a flaw reported today poses a far greater immediate risk than one reported last year. She finds negotiations with vendors increasingly unmanageable, particularly in determining whether a fix is effective or if a grace period should apply.
Chris Fistonich believes that the discovery of critical vulnerabilities, once slow and costly, has become rapid with AI, while defenders are falling behind in patching. He considers it likely that this situation will persist. Calif, for its part, advocates for reducing unused code and dependencies, a process it calls "uncoding of vibe." The company contests the usefulness of maintaining a manually written Base64 decoder in assembly to handle a few hundred bytes.
Brief IA — L'actualité IA en français
L'essentiel de l'actualité de l'intelligence artificielle, décrypté et expliqué chaque jour.