Executive brief
The SageMaker Python SDK, used for building and deploying machine learning models on AWS, contains two security vulnerabilities. One flaw allows users with specific permissions to steal secret keys and tamper with data stored in Amazon S3, while the other disables secure encrypted connections (SSL/TLS), potentially allowing attackers to intercept or modify data during model downloads. These issues could lead to unauthorized data modification or the compromise of sensitive machine learning assets.
Technical details
Two vulnerabilities were identified in the SageMaker Python SDK. CVE-2026-1777 involves the exposure of per-job HMAC keys through environment variables accessible via the DescribeTrainingJob API; an attacker with sufficient IAM permissions can retrieve these keys to forge serialized payloads and overwrite S3 objects. CVE-2026-1778 is a flaw where SSL certificate verification was globally disabled in the Triton Python backend to bypass errors during model downloads, exposing all HTTPS traffic to man-in-the-middle (MitM) attacks. The HMAC issue is resolved in versions 2.256.0 and 3.2.0, while the TLS issue is resolved in 2.256.0 and 3.1.1.
Affected products
- AWS SageMaker Python SDK v2 < v2.256.0, v3 < v3.2.0
CVE identifiers
- CVE-2026-1778
- CVE-2026-1777
Timeline
- 2026-02-02: disclosed
- 2026-02-02: advisory