AI engineering glossary · Evals guardrails AI security
What is misinformation risk?
Misinformation risk: An OWASP-listed concern that models produce false or misleading content that users or downstream systems trust, including hallucinated facts and biased answers. Overreliance on such output can cause real harm in medicine, law and finance, so verification and disclosure are needed.
Misinformation risk: an example
A health chatbot confidently recommends a wrong dosage and a user acts on it without checking a doctor.
Related terms
More Evals guardrails AI security terms
- MITRE ATLAS
- MMLU
- Model extraction
- Model inversion
- Model signing
- Model supply chain attack
- Multimodal prompt injection
- NeMo Guardrails
Where you learn it
Misinformation risk is part of the AI Engineering Program, which teaches it with notes, a hands-on lab and an exam. The first lesson is free.