Junglewise Threat Intelligence

CVE-2026-10766: MLRun weak hash in calculate_dataframe_hash

CVE-2026-10766 · Severity: low · CVSS 3.6 · Published 2026-06-03

Executive brief

MLRun, an open-source MLOps platform, is affected by a vulnerability where different datasets can be assigned the same unique identifier (hash). This can lead to data corruption or path conflicts where one dataset accidentally overwrites another. In a shared environment, this could allow a user to unintentionally or maliciously interfere with the integrity of machine learning models and their associated data.

Technical details

A vulnerability in `mlrun.utils.helpers.calculate_dataframe_hash` (CWE-327/CWE-328) stems from the use of a weak hashing implementation that relies on `pandas.util.hash_pandas_object`. This implementation is susceptible to collisions across different DataFrames, particularly involving schema/dtype differences (e.g., bool vs int) and cross-column hash combinations. An attacker with local environment access could potentially craft DataFrames that resolve to the same artifact path, leading to integrity and availability impacts. A proposed fix involves migrating to SHA-256 and incorporating stable DataFrame schema metadata, index labels, and dtypes into the hash calculation.

Affected products

  • Iguazio mlrun <= 1.12.0rc3

Timeline

  • 2026-05-16: disclosed: Issue and Pull Request opened on GitHub
  • 2026-06-03: advisory: GitHub Advisory and NVD entry published

References

Related threats