OWASP LLM03 (2023) Training Data Poisoning

What is OWASP LLM03 (2023) Training Data Poisoning?

Training data poisoning targeted the datasets used to build or adapt a model - introducing trigger behaviors or systematic bias that only show up under specific conditions.

What this looks like in production

OWASP broadened this in 2025 to LLM04 Data and Model Poisoning, explicitly including RAG indexes and runtime knowledge bases, not only classical training pipelines.

Further reading on Giskard

Official OWASP reference

Get AI security insights in your inbox