Calif, a security research firm, disclosed Java4Shell (CVE-2026-114938) on November 17, 2026: a memory-corruption flaw in OpenJDK’s Base64 decoder that it says can lead to remote code execution. Calif scores it 9.2 of 10 under CVSS 4.0. Its write-up says one researcher found the flaw with OpenAI’s Codex in an hour, and adds: “The total token cost was less than $200.”
The disclosure landed on the same day as a Forbes report that OpenAI’s GPT-6 Cyber has been finding flaws across Apple products. This page records what Calif published and what the reports say, and keeps the two apart.
On this page · 8 sectionsOpen
Calif disclosed Java4Shell (CVE-2026-114938), a memory-corruption flaw in OpenJDK’s Base64 decoder, on November 17, 2026.
It scores 9.2 of 10 under CVSS 4.0 and needs an x86-64 processor with AVX-512 VBMI; ARM64 CPUs are not affected.
Calif says its researcher found it with Codex in an hour and built a first exploit in two more, for under $200 in tokens.
Oracle released a patch at 1:00pm PT on November 17, and Calif says all Java users should update immediately.
Forbes reported the same day that OpenAI’s GPT-6 Cyber has been finding flaws across Apple devices.
§ 01What Java4Shell is
| Item | What Calif says |
|---|---|
| Name and ID | Java4Shell, CVE-2026-114938 |
| Where | OpenJDK’s Base64 decoder, in hand-written assembly for x86-64 |
| Severity | 9.2 of 10 under CVSS 4.0, critical |
| Affected | OpenJDK from version 17.0.6 or newer |
| Hardware | x86-64 processors with AVX-512 VBMI; ARM64 CPUs are not affected |
| Trigger | 64 A characters and one equals sign (128 on older OpenJDK 17) |
| Found by | Calif’s Quang Luong, working with Codex |
| Fix | OpenJDK advisory and Oracle patch, 1:00pm PT on November 17, 2026 |
The bug sits in the code that handles the last few characters. For a 65-character input, OpenJDK sets aside a 50-byte result array, then miscalculates a negative length: “That invalid length becomes a full 64-byte write, sending 62 bytes past the end of the array.” The input is still refused, but too late: “Java eventually rejects the input, but by then neighboring objects may already be corrupted.” Calif says that corruption can leak heap memory, as Heartbleed did, or be chained into remote code execution.
§ 02Where Calif showed it working
Calif says it demonstrated remote code execution in these applications, in their default settings:
| Before login | After login |
|---|---|
| Keycloak | Apache Kafka Broker |
| Atlassian Jira | Apache Solr |
| Atlassian Confluence | Elasticsearch |
| JFrog Artifactory | Gerrit |
| Jenkins | Spinnaker |
| PingFederate |
- Not a full list. Calif calls both lists non-exhaustive and says it also reached code execution in sample apps built on Apache Struts and Spring Boot.
- Clients too. It adds that a Java program opening a TLS connection to a malicious server can be compromised, from an exploit shared by the researcher samczsun.
- No single exploit. Calif says each application needs its own steps, which its team built with AI agents after setting up pipelines first.
§ 03How AI found it
Calif’s researcher Quang Luong describes the start plainly: “Well, more precisely, I told Codex to look at those areas like any serious security researcher in 2026.” Most early findings, he writes, were unusable, until Codex returned the Base64 bug. The firm’s summary: “It took our own Quang Luong an hour to discover the vulnerability, and two more hours to develop the first working exploit.”
Calif frames this as the era of the “centaur hacker”, part human and part AI: “AI brings the horsepower. Our researchers bring the context.” It ends with a proposal it calls vibe-uncoding, using the same tools to find and delete code a system does not need.
Calif’s write-up names the tool, Codex, and not the model behind it. Techmeme’s summary of the New York Times report reads: “Cybersecurity startup Calif says it used GPT-6 Cyber to uncover an RCE flaw in Java”. Calif had said earlier in the day on X, at 16:11 UTC: “Since the vulnerability has leaked: Calif discovered and reported Java4Shell (CVE-2026-114938).” That post had about 175,000 views when we read it at 22:19 UTC.
§ 04The Apple report
Forbes reporter Thomas Brewster posted his story on X at 19:32 UTC: “OpenAI’s new hacking model GPT-6 Cyber has been finding a load of vulnerabilities in Apple technologies.” He added: “Not just iPhones and Macs, but also Vision Pro, TV and Watch.” A post by Watcher Guru at 19:26 UTC, with about 737,000 views by 22:19 UTC, summarized it as “New OpenAI model discovers 35 vulnerabilities in Apple devices”, with flaws that could let attackers take over iPhones and Macs.
We did not read the Forbes article, and we have not seen a public list of the Apple flaws or Apple’s account of them, so the count of 35 here is Watcher Guru’s summary of Forbes, not a figure we checked. Our record of GPT-6 Cyber covers what OpenAI said the model is for and who can use it.
§ 05What to do
- Update Java. Calif’s advice is short: “All Java users should update immediately.” Oracle lists 8u511b33-perf, 17.0.21.0.3, 21.0.13.0.3, 25.0.5.0.3 and 27.0.1.0.3 among the vulnerable Oracle Java SE versions, per the write-up.
- Check the hardware. The flaw needs an x86-64 processor with AVX-512 VBMI; Calif’s write-up includes a check for servers.
- Update Apple devices. If the Apple report holds, the fixes arrive through normal Apple updates.
§ 06What is not established
- Which model found Java4Shell. Calif names Codex; the GPT-6 Cyber attribution comes from the New York Times as summarized by Techmeme.
- The Apple list. No public list of the Apple flaws has reached us.
- Use in the wild. Calif’s write-up reports no attacks using the flaw; it says details leaked before publication.
§ 07Where CellCog sits
We build CellCog, a platform where a business hires AI employees, and those employees run on Anthropic’s Claude models. CellCog routes no work to GPT-6 Cyber or Codex, and has no tie to Calif. Our AI software engineer works in a customer’s codebase behind an approval step, so no command reaches a customer’s systems unclassified.
§ 08Sources
- Calif, Java4Shell: Write once, RCE everywhere, November 17, 2026, read at 22:13 UTC
- Calif on X, 16:11 UTC, and the write-up post, 21:11 UTC, November 17, 2026
- Thomas Brewster on X, sharing the Forbes report, 19:32 UTC; the Forbes article (linked, not read)
- Watcher Guru on X, 19:26 UTC
- Techmeme on X, summarizing the New York Times, 21:21 UTC
Q1What is Java4Shell?
Java4Shell (CVE-2026-114938) is a memory-corruption flaw in OpenJDK’s Base64 decoder, disclosed by Calif on November 17, 2026. Calif scores it 9.2 of 10 under CVSS 4.0 and says it can lead to remote code execution.
Q2Which systems are affected?
Calif says OpenJDK from version 17.0.6 or newer, on x86-64 processors with AVX-512 VBMI support. ARM64 CPUs are not affected. Oracle OpenJDK, Oracle Java SE and Oracle GraalVM for JDK are listed.
Q3Has Java4Shell been patched?
Yes. Calif says OpenJDK released a security advisory and Oracle released a patch at 1:00pm PT on November 17, 2026, and that all Java users should update immediately.
Q4Was Java4Shell found by AI?
Calif’s write-up says its researcher Quang Luong told Codex to audit parts of OpenJDK, found the flaw in an hour and built a first exploit in two more hours, for under $200 in tokens. Techmeme’s summary of the New York Times report says Calif used GPT-6 Cyber.
Q5What did GPT-6 Cyber find in Apple products?
Forbes reported on November 17, 2026 that OpenAI’s GPT-6 Cyber has been finding vulnerabilities across Apple products, including iPhones, Macs, Vision Pro, TV and Watch. A widely shared post put the count at 35; we have not seen a public list.
