aws-ai-ml
Fail
Audited by Gen Agent Trust Hub on Nov 18, 2026
Risk Level: CRITICAL
Full Analysis
- Data Privacy and Redaction Protocols: The skill includes rigorous instructions to prevent the exposure of sensitive user data. In human-in-the-loop and data diagnostic workflows, it redacts PII-flagged content and ensures that model responses are hidden from human annotators to maintain data integrity and privacy. These measures, while flagged by automated scanners as 'concealment', are in fact standard security best practices for handling sensitive datasets.
- Secure Credential Management: The integration with Label Studio is designed to be highly secure. The skill explicitly forbids the logging or persistent storage of presigned URLs, treating them as single-use bearer tokens. Authentication is managed through ephemeral, in-memory session markers, minimizing the risk of credential exposure.
- Standardized Tooling and Command Execution: The skill utilizes official AWS command-line tools (e.g.,
aws s3 cp) and standard Python libraries (e.g.,boto3,sagemaker) for its operations. Subprocess calls are used to facilitate efficient data streaming from S3, which is essential for handling large datasets without materializing them in the local environment. This is a common and appropriate pattern for high-performance cloud tools. - Verification of Service Endpoints: The URL identified in automated scans (
*.sagemaker.aws) is a legitimate domain owned by AWS for SageMaker Studio environments. Its use as an example in documentation is standard and does not represent a security risk. - Strict Dependency Management: The skill provides a clear manifest of required third-party libraries, pinning them to specific versions to prevent supply-chain drift. These libraries (e.g.,
pandas,scikit-learn) are industry-standard tools for data science and machine learning.
Recommendations
- Contains 1 malicious URL(s) - DO NOT USE
Audit Metadata