AI Dictionary › AI Fundamentals

Data poisoning

Data poisoning (avvelenamento dei dati)

Data poisoning is an attack in which someone deliberately inserts corrupted or deceptive data into the material used to train a model, so as to manipulate its behavior. It is like sabotaging the library a student studies from: slip in books with targeted errors and the student will learn those falsehoods and repeat them with conviction. With AI, polluting even a small fraction of the data can install a "backdoor": a hidden trigger, such as a specific word or symbol, that fires unwanted behavior only when it appears. The model seems normal otherwise, which makes the attack hard to detect.

Definition

Data poisoning matters because much training data comes from open, unverified sources like the web. A patient adversary can contaminate them in advance, and the flaws stay invisible until triggered, undermining trust in the whole data pipeline.

Related terms

• Red teaming
• Jailbreak
• AI Bias
• Machine Learning
• Guardrails

More in AI Fundamentals

• LLM (Large Language Model)
• Token
• Context Window
• Temperature
• AI Hallucination
• Fine-tuning
• RAG (Retrieval Augmented Generation)
• Embedding

Put it into practice

• Real scenario: apply this concept with an instant AI verdict.
• Daily gym: short drills to train dictionary concepts.
• Playoff game: challenge other players, elimination-style.

📱 Download the Android app (beta) iOS coming soon

Say what you mean. Get what you need.

Grace Certified, the AI coach that trains and certifies your prompt engineering, by Agora Intelligence.