The vulnerability arises from the mlrun.utils.helpers.calculate_dataframe_hash function in mlrun’s DataFrame Hash Handler, where a weak hash algorithm is employed. This flaw enables an attacker to craft input that may produce hash collisions, undermining the integrity checks that rely on these hashes. While the description suggests that the exploit is complex and difficult, the use of an insecure hash could still allow a local adversary to bypass integrity verification or substitute data without detection.

mlrun versions up to 1.12.0-rc3 are affected. The vulnerability is confined to environments where mlrun is installed locally, and no remote attack vector is documented.

The CVSS score of 2.0 categorizes the risk as low and the EPSS score is unavailable, indicating limited public exploitation data. The exploit is limited to a local environment, requires high complexity, and is noted as difficult to execute. It is not listed in the CISA KEV catalog, further reducing the immediate threat level, but the weakness presents a long‑term concern for data integrity should the hash function be relied upon for security-critical decisions.