Executive brief
AWS SageMaker's Python SDK uses MD5 hashing to track and cache machine learning workflow configurations. Due to MD5's known collision vulnerability, attackers or accident can cause different workflow configurations to produce identical hashes, leading to incorrect workflow reuse, data processing errors, and pipeline integrity violations that may go undetected.
Technical details
The vulnerability is a weak cryptographic hash function issue (CWE-328, CWE-440) in the SageMaker Workflow component where MD5 is used for configuration fingerprinting and caching. MD5's collision resistance is broken, allowing two different workflow configurations to hash to the same value. This causes workflows to be inadvertently replaced or reused with incorrect parameters, compromising pipeline integrity. The attack requires only the ability to submit or modify workflow configurations—no authentication bypass or network exploit is needed. AWS has patched this in version 2.237.3 and later by replacing MD5 with SHA-256 for file hashing.
Affected products
- AWS SageMaker Python SDK all versions before 2.237.3
Timeline
- 2025-03-20: disclosed: GHSA-32g6-mg92-ghm2 and CVE-2025-0508 published
- 2025-03-20: patched: Fix available in version 2.237.3 and later; commit dcdd99f replaces MD5 with SHA-256