Local file disclosure via dereferenced HDF5 links
Published Aug 2, 2026 · Updated Aug 2, 2026
HDF5 link handling in Keras 3.14.0 and earlier allows remote attackers to disclose local HDF5 file contents through a crafted model file. KerasFileEditor and keras.saving.load_weights use raw HDF5 subscripts that automatically follow attacker-controlled ExternalLink or SoftLink targets instead of safe link-checking helpers. Opening a malicious .h5, .weights.h5, or .keras file is required; readable model weights and other HDF5 datasets on the local host can enter editor state or a loaded model.
Summary
What happened
HDF5 link handling in Keras 3.14.0 and earlier allows remote attackers to disclose local HDF5 file contents through a crafted model file. KerasFileEditor and keras.saving.load_weights use raw HDF5 subscripts that automatically follow attacker-controlled ExternalLink or SoftLink targets instead of safe link-checking helpers. Opening a malicious .h5, .weights.h5, or .keras file is required; readable model weights and other HDF5 datasets on the local host can enter editor state or a loaded model.
The record
- CVE
- CVE-2026-9335
- Published
- Aug 2, 2026
- Updated
- Aug 2, 2026
- Vendor
- Keras
- Product
- Keras
- Classifications
- CWE-22, T1204.002, T1005
- Attack vector
- network
- Privileges
- unauthenticated
Timeline
How it unfolded
- Aug 2, 2026CVE publishedPublication date reported by the CVE source.
- Aug 2, 2026Record updatedLatest update available in the CVE record.
Exploitability
Present is not the same as exploitable
Compare your product and version with the public record. A matching version still requires validation against your environment.
Is a vulnerable build present?
Compare these published version ranges with your installed build and any vendor patches.
- Affected versionversion=unspecified <3.12.3, 3.15.0
What conditions does exploitation require?
What is affected?
Published CVSS scores
CVSS describes severity. EPSS estimates exploitation probability.
Attacks
What attackers are doing with it
Daily unique IPs observed by Shadowserver honeypots for known exploited vulnerabilities (KEVs). Missing observations do not establish an absence of attacks.
Weakness, pattern, technique
Public exploit references
- KerasFileEditor ExternalLink regression demonstrationproof of concept · demonstrated
Labels summarize the accepted research assessment. They do not indicate a test against your environment.
Technologies
Your stack
See the directory against your own environment.
Your stack
Check the software in your environment
Book a demo to see how Hinoki identifies affected software and validates exploitability in your environment.
Book a demo