← Blog
JavaOraclePre-auth RCEPatch Management

CVE-2026-114938: Oracle confirms Java flaw at 8.1, not 9.2

Oracle confirmed CVE-2026-114938 in its November CSPU: a Java SE Compiler flaw scored 8.1, not the rumored 9.2 — difficult to exploit, no public PoC.

Zero Hunt Research··7 min read

Published by Zero Hunt, an autonomous AI red team on an on-premise appliance running private AI: automated penetration testing for networks and infrastructure, black-box or gray-box, with a human approving every step that matters.

Today Oracle shipped a Critical Security Patch Update that fixes 11 Java SE vulnerabilities, every one of them remotely exploitable without authentication. In parallel a louder number is circulating: a reported CVSS 9.2 pre-authentication remote code execution in Java 17 and above, tracked as CVE-2026-114938. The 9.2 is the headline. It is also, as of tonight, unconfirmed by any primary source — and that gap is the story.

Developing story — first published 20:58 CEST (19:58 UTC), 2026-11-17. Updated as Oracle, OpenJDK and the NVD publish more.

Update — 2026-11-18 03:16 CEST

Oracle has now confirmed CVE-2026-114938. It published the CVE with the November 2026 Critical Security Patch Update, and the NVD record (source: Oracle, published 2026-11-17 22:18 UTC) settles the questions this post opened with:

  • It is real, and it is one of the 11 Java SE flaws in today's CSPU — not a separate, mysterious bug. The affected component is the Compiler.
  • The score is 8.1 (CVSS 3.1), not the rumored 9.2. Oracle, the primary CNA, scored it CVSS:3.1/AV:N/AC:H/PR:N/UI:N/S:U/C:H/I:H/A:H — "difficult to exploit" (high attack complexity), unauthenticated, network-reachable, full confidentiality/integrity/availability impact. The "9.2 (v4.0)" that circulated still has no authoritative source behind it.
  • No known exploitation and no public PoC. CISA's SSVC decision on the NVD record is exploitation: none, automatable: no (2026-11-17 21:53 UTC), and it is not in CISA KEV.

The caution this post argued for held: the headline number was wrong, the real flaw is a difficult-to-exploit entry at the top of a routine patch batch, and the right move was always to patch the confirmed CSPU set rather than scramble over a floating CVSS. The At-a-glance box below now reflects the confirmed facts; the analysis sections are kept as first published, with their original timestamps.

At a glance

CVE CVE-2026-114938
Product / affected versions Oracle Java SE 8u511b33-perf · 17.0.21.0.3 · 21.0.13.0.3 · 25.0.5.0.3 · 27.0.1.0.3; Oracle GraalVM for JDK 17.0.21.0.3 · 21.0.13.0.3; GraalVM EE 21.3.20.0.3 (component: Compiler)
Fixed in Oracle Critical Security Patch Update, November 2026 (released 2026-11-17)
CVSS 8.1 (CVSS 3.1), High — scored by Oracle (Primary CNA); CVSS:3.1/AV:N/AC:H/PR:N/UI:N/S:U/C:H/I:H/A:H. The "9.2 (v4.0)" that circulated has no authoritative source
Exploited in the wild No; CISA SSVC exploitation: none on the NVD record (2026-11-17)
CISA KEV Not listed (catalog version 2026.11.17)
Public PoC None public as of 2026-11-18
Official advisory Oracle CSPU November 2026

The reported Java 17 pre-auth RCE, and what backs it

On November 17, Rapid7's Emergent Threat Response team wrote that "public reports described CVE-2026-114938 as a memory corruption vulnerability in Java 17 and above that could allow a remote attacker to execute code without authentication," with a "reported severity ... Critical, with a CVSSv4 score of 9.2." The qualifier in the same post is the sentence that matters: "These CVE-specific details remain preliminary, pending confirmation in Oracle or OpenJDK advisories."

As of this writing (the night of November 17), that confirmation does not exist. The NVD has no record of CVE-2026-114938 — its API returns zero results for the ID. It is not in the CISA Known Exploited Vulnerabilities catalog (version 2026.11.17). OpenJDK's vulnerability advisory index lists nothing newer than October 24, 2026. No vendor has named a component, an affected build, a fix, or a proof of concept. The "9.2" has one publicly identifiable origin: an unnamed report relayed by a single vendor blog. (This resolved hours later — see the Update above: Oracle published the CVE at 8.1, Compiler component, as part of the CSPU.)

That does not make it false. Memory-corruption RCEs in the JDK are real and have shipped before. It makes it unverified, which is a different operational category — and it means the number should not, on its own, trigger an emergency change window.

The Java flaws Oracle actually shipped today

What is confirmed deserves more of your attention tonight. Oracle's Critical Security Patch Update for November 2026 — a new "targeted, high-priority" format that Oracle says complements its "existing quarterly cumulative Critical Patch Updates" — released on Tuesday, November 17, 2026.

Its Java SE section is blunt: "This Critical Security Patch Update contains 11 new security patches for Oracle Java SE. All of these vulnerabilities may be remotely exploitable without authentication." The highest CVSS v3.1 base score across those 11 is 8.1 — not 9.2. That 8.1 flaw is now known to be CVE-2026-114938 itself (component: Compiler), which Oracle confirmed hours after this post went live. The affected products and baselines Oracle lists:

  • Oracle Java SE 8u511, 8u511-perf, 11.0.33, 17.0.21, 21.0.13, 25.0.5, 27.0.1
  • Oracle GraalVM for JDK 17.0.21, 21.0.13; GraalVM Enterprise Edition 21.3.20.0.3
  • Oracle JavaFX 21.0.13, 25.0.5, 27.0.1

Eleven network-reachable, no-credentials-required flaws in the most widely deployed runtime in the enterprise is not a quiet update. If you run server-side Java that terminates untrusted input — application servers, message brokers, middleware, anything parsing serialized objects, fonts, images or scripts off the network — this CSPU is the thing to schedule.

Remediation

  1. Am I affected? Run java -version on every host, and — more importantly — inventory the JVMs embedded inside applications (app servers, agents, bundled runtimes) that a shell java -version never shows. Any build at or below Oracle's baselines above (8u511, 11.0.33, 17.0.21, 21.0.13, 25.0.5, 27.0.1, or the matching GraalVM/JavaFX trains) is in scope for the 11 confirmed flaws, CVE-2026-114938 among them. Rank the JVMs that accept untrusted network input to the top.
  2. Patch the confirmed set. Apply the November 2026 CSPU for your JDK train; the fixed builds ship with it, and they fix CVE-2026-114938 along with the other ten. These 11 are remotely exploitable without authentication — the profile you treat as patch-now. Read the published Java SE risk matrix for the exact fixed builds and CVE IDs.
  3. Can't patch immediately? Reduce what the network can reach: put internet-facing Java services behind authenticated proxies, close management and debug ports, and filter the endpoints that deserialize or render untrusted content. The high attack complexity Oracle assigned (AC:H) means exploitation is non-trivial, which buys scheduling room — it is not a fix.
  4. Hunt on evidence, not on invention. There are no vendor or researcher IOCs for CVE-2026-114938 — it is a difficult-to-exploit flaw with no observed exploitation, so no indicators exist yet; treat anything published as an IOC for it as a guess. For any pre-auth flaw in this batch, watch Java-facing services for the ordinary tells of a network RCE: a JVM spawning unexpected child processes (java → sh/cmd), outbound connections from a host that normally only serves, and crashes or restarts clustered on one endpoint (MITRE ATT&CK T1190 Exploit Public-Facing Application, T1059 Command and Scripting Interpreter).
  5. Eradicate and verify. After patching, confirm the running build actually changed — restart the service, because a patched jar on disk with the old JVM still resident is not fixed — and rotate credentials reachable from any host you believe was exposed. For a difficult-to-exploit Compiler flaw like this one, the useful verification is whether the vulnerable code path is reachable and drivable on your exact build, not simply whether the version string sits below a baseline.

What we now know

Oracle's confirmation (see the Update above) closes the questions this post opened with:

  • CVE-2026-114938 is real and distinct — an Oracle Java SE Compiler flaw, published by Oracle as the CNA.
  • Scope: Oracle Java SE 8u511b33-perf, 17.0.21.0.3, 21.0.13.0.3, 25.0.5.0.3, 27.0.1.0.3, plus the matching GraalVM for JDK and GraalVM Enterprise Edition trains. Fixed by the November 2026 CSPU.
  • The real CVSS: 8.1 (v3.1), "difficult to exploit," scored by Oracle. The "9.2 (v4.0)" never had an authoritative source.
  • Exploitation: none known; CISA SSVC records exploitation: none, and it is not in CISA KEV.

What is still outstanding is the full per-CVE risk matrix for the other ten flaws in the batch, which Oracle publishes alongside the CSPU advisory.

Patch the proven set; validate the rumor

When a critical-sounding number appears with no vendor behind it, the reflex to patch everything and the reflex to wait both cost you. The useful move is neither — it is to establish whether the thing is real against your own systems. That is the core argument for an autonomous AI red team that runs on-premise on its own private models, with a human approving each step: it can fingerprint the exact JDK build behind each service and, gray-box, try to actually reach and drive the reported code path, so your response is governed by whether the flaw is exploitable on your estate rather than by a floating CVSS score — the same discipline that separates a validated finding from a headline in automated penetration testing. For the 11 flaws Oracle confirmed, the AI Remediation Advisor ranks them by real exploitability — CISA KEV first, then what a campaign proved reachable, then CVSS, then EPSS, in the spirit of risk-based patching under BOD 26-04 — and returns the exact fixed build with rollout and rollback notes, re-verifying each fix with the same proof that demonstrated the issue.

Every vulnerability CISA lists as exploited, with federal due dates: CISA KEV tracker →

Patching one CVE at a time? A CTEM program validates which exposures are actually exploitable before the next advisory: what CTEM is →

Is this exploitable in your environment?

Zero Hunt answers that on your own network: an autonomous AI red team on an on-premise appliance, running on private AI, black-box or gray-box, with a human approving every step that matters. Proof of what is exploitable, the fix, and signed evidence — no data leaves your perimeter.