Browse / Data and Model Poisoning
Data and Model Poisoning
Tampering with training data or model weights, including backdoors. 4 records. Reference: OWASP LLM04 · ATLAS AML.T0020.
- Other weaknesses 1
- Authentication and access control 1
- Training Data Poisoning 1
- Manipulate AI Model: Poison AI Model 1
- NLTK downloader allows cross-package resource and model poisoning
- MLflow: Deterministic sampling in dataset digest enables predictable collisions Dataset digests use deterministic sampling with a weak hash, so an attacker can craft a different dataset that produces the same digest.
- AI Model Tampering via Supply Chain Attack
- Web-Scale Data Poisoning: Split-View Attack