Security firm Calif has disclosed Java4Shell, a flaw it says can let attackers run their own code on vulnerable Java systems without credentials. Oracle released a patch on November 17, 2026, according to Calif and CyberScoop. The flaw, tracked as CVE-2026-114938, affects OpenJDK, a widely used Java runtime—the software that runs Java applications. Source Source
Calif says exposure requires an affected runtime, an x86-64 processor with AVX-512 VBMI support and an application path that lets attacker-controlled data reach OpenJDK’s Base64 decoder. ARM64 CPUs are unaffected. Exploitation also needs application-specific preparation: “There is no universal payload,” the firm cautions. Source
A decoder with boundary issues
The bug sits in OpenJDK’s optimized Base64 decoder. Base64 represents binary data as text and appears in authentication protocols, file uploads and network messages. On supported processors, OpenJDK uses specialized assembly code to speed up decoding—and Calif found a memory-corruption flaw in that fast path. The firm rates it critical, at 9.2 out of 10 under CVSS 4.0. Source
According to Calif, malformed padding causes the decoder to calculate an invalid length and write beyond its allocated memory. Java then rejects the bad input, but neighboring objects may already be damaged. Rejecting invalid data is a fine idea. Doing it after the memory overwrite is less helpful. With careful exploitation, Calif says, that corruption can expose sensitive memory or lead to attacker-controlled code execution. Source
In its testing, Calif says it achieved code execution without credentials in default configurations of Keycloak, Jira, Confluence, Artifactory, Jenkins and PingFederate. Its demonstrations involving Kafka Broker, Solr and Elasticsearch required authentication. Source
Calif has published a proof of concept for the corruption bug, but says it is withholding application-specific exploit chains to give users time to upgrade. Turning that public trigger into a server takeover still requires a tailored exploit. Source
What the AI actually did
Calif says researcher Quang Luong found the vulnerability in about an hour while auditing OpenJDK with OpenAI’s Codex, then spent two more hours developing the first working exploit. It puts the total token spend below $200. That is a token bill, not a complete research budget. Source Source
Calif says most early AI findings were unusable. Its researchers built tools to help agents understand memory layout and established exploitation pipelines and strategies. Most later exploit chains were then built almost autonomously with AI, the firm says. Source
Patch the runtime you actually run
Calif urges immediate updates. Other OpenJDK vendors release on their own schedules; check your distributor’s CVE-specific guidance to confirm its fix is available. Orca recommends inventorying Java installations—including runtimes bundled inside applications and container images—and prioritizing internet-facing services that decode untrusted Base64 data. Source Source Source
CyberScoop said there were no reports of exploitation in the wild as of its article’s publication. An exposed system still belongs in the patch queue. Source Source
Comments