## Summary
A security flaw in h2oai h2o-3 (up to 7402) allows remote attackers to trigger deserialization via the importBinaryModel function in h2o-core/src/main/java/hex/Model.java (JAR Handler). The exploit is reported as publicly released, and the vendor did not respond after early contact.
## Technical Details
- Vulnerable component: JAR Handler in h2o-core/src/main/java/hex/Model.java
- Vulnerable function: importBinaryModel
- Issue type: deserialization triggered by manipulation of input processed by the model import path
- Attack vector: remote/network reachable (AV:N), requiring no privileges (PR:N) and no user interaction (UI:N)
- Public exploit availability: the NVD notes the exploit has been released to the public
## Affected
- h2oai h2o-3 versions up to 7402 (listed as h2o <= 7402)
## Impact
- Remote deserialization can occur, with confidentiality, integrity, and availability impact indicated by the CVSS (C:L, I:L, A:L).
- Publicly available exploit increases likelihood of real-world attacks.
## Mitigation
1. Patch immediately to a version that addresses the deserialization issue in importBinaryModel (no fixed version details were provided in the available material).
2. Restrict network access to H2O endpoints/services that accept model uploads/imports, since the attack is remote (AV:N).
3. Monitor for suspicious model import attempts consistent with manipulated JAR/model inputs (specific detection indicators were not provided in the available material).
CVSS:3.1/AV:N/AC:L/PR:N/UI:N/S:U/C:L/I:L/A:L
CWE-20