model-deployment
Warn
Audited by Gen Agent Trust Hub on Feb 17, 2026
Risk Level: MEDIUMREMOTE_CODE_EXECUTIONCOMMAND_EXECUTION
Full Analysis
- Dynamic Execution (MEDIUM): The code uses
joblib.load('model.pkl')to load a pre-trained model. Joblib relies on Python's pickle module, which is inherently insecure and can execute arbitrary code during deserialization. Loading models from untrusted sources is a significant security risk. - Command Execution (LOW): The deployment instructions include shell commands for
dockerandkubectl. These operations require high privileges and could be used to compromise the container registry or Kubernetes cluster if instructions were maliciously modified. - Indirect Prompt Injection (LOW): The FastAPI server defines a prediction endpoint that ingests external data. This creates an attack surface for untrusted data to enter the system context. 1. Ingestion points:
PredictionRequestinapp.py. 2. Boundary markers: None. 3. Capability inventory:joblib.load,docker,kubectl. 4. Sanitization: Type checking via Pydantic.
Audit Metadata