PatchSiren

PatchSiren cyber security CVE debrief

CVE-2026-8751 H2o CVE debrief

CVE-2026-8751 describes a remotely reachable flaw in h2oai h2o-3 up to 7402, centered on importBinaryModel in h2o-core/src/main/java/hex/Model.java and the JAR handling path. The issue is described as a manipulation that leads to deserialization, and the CVE notes that exploit code has been made public. For defenders, this is most important anywhere binary model imports are exposed to untrusted input or reachable over the network.

Vendor
H2o
Product
Unknown
CVSS
MEDIUM 5.5
CISA KEV
Not listed in stored evidence
Original CVE published
2026-05-17
Original CVE updated
2026-05-19
Advisory published
2026-05-17
Advisory updated
2026-05-19

Who should care

Security teams, platform owners, and MLOps operators running h2oai h2o-3 up to 7402, especially if binary model import features are exposed to users, APIs, automation, or other untrusted sources.

Technical summary

The supplied CVE record and NVD metadata point to a remote deserialization weakness in the importBinaryModel path of h2o-core/src/main/java/hex/Model.java, associated with JAR handling. The vulnerability is mapped to CWE-502 and CWE-20 in the source metadata, and the CVSS vector indicates network attackability with no privileges or user interaction required. The CVE description also states that public exploit code exists, which raises operational risk even though no remediation details are provided in the corpus.

Defensive priority

Medium severity on paper, but higher operational priority if the affected import path is reachable from untrusted sources or any internet-facing service. Public exploit availability and lack of vendor response increase the need for immediate containment and monitoring.

Recommended defensive actions

  • Inventory h2oai h2o-3 deployments and confirm whether any instance is running version 7402 or earlier.
  • Identify whether binary model import or JAR handling paths are exposed to untrusted users, APIs, jobs, or automation.
  • Restrict access to model import workflows, and avoid accepting untrusted model files from unauthenticated or low-trust sources.
  • Apply compensating controls such as network segmentation, strong authentication, least privilege, and sandboxing around model ingestion.
  • Monitor for unexpected model import activity, deserialization-related errors, crashes, or other anomalous behavior on affected systems.
  • Track the CVE, NVD, and vendor references for any fix or mitigation guidance and plan an upgrade or other corrective action when available.

Evidence notes

This debrief is based on the supplied CVE description, NVD metadata, and the referenced VulDB CNA-linked sources. The corpus states h2oai h2o-3 up to 7402, identifies importBinaryModel in h2o-core/src/main/java/hex/Model.java, and says the attack can be carried out remotely with public exploit code. Source metadata also lists CWE-20 and CWE-502. Vendor identity in the enrichment data is low-confidence/needs review, so the debrief avoids asserting more than the supplied evidence supports.

Official resources

Publicly disclosed on 2026-05-17. The supplied description says exploit code has been released publicly and that the vendor was contacted early but did not respond.