Cybersecurity glossary
What is data poisoning?
Also known as: model poisoning
Definition
Data poisoning is the manipulation of data used to train, fine-tune or ground an AI model so that it behaves in a way the attacker wants.
Why it matters
Effects range from subtle bias and wrong answers to hidden backdoors triggered by specific inputs. In retrieval-augmented systems, poisoned documents in a knowledge base can steer answers at run time.
Track data provenance, limit who can add content, and re-test model behavior after data changes.
Common questions about data poisoning
See also
Sources
Related Guides
GuideGRC
10-Point GDPR Compliance Checklist for Security and Privacy Teams
A practical GDPR checklist covering data mapping, lawful basis, records of processing, PII protection, DPIAs, processor contracts, data subject rights, international transfers and 72-hour breach notification.
Read more GuideSOC
How to Build an Agentic SOC: A 4-Phase Roadmap for AI Security Operations
A practical roadmap for moving a SOC to AI agents: build the data foundation, start with quick wins, customize and orchestrate agents, and keep humans in command.
Read more