Caesar AI Atlas

Data Poisoning (Datenvergiftung)

Also known as: Poisoning

Caesar AI Atlas Definition

Data Poisoning ist ein adversarialer Angriff, bei dem ein böswilliger Akteur in die Daten oder den Trainingsprozess eingreift, die zum Aufbau eines Modells verwendet werden. Durch Einfügen, Verändern oder Korrumpieren von Trainingsbeispielen versucht der Angreifer, Modellverhalten zu manipulieren, Leistung zu verschlechtern oder gezielte Ausfälle zu erzeugen.

Other Definitions

Data Poisoning (Datenvergiftung) Source

A type of cyberattack in which a malicious actor intentionally compromises a training dataset used by an AI model to influence or manipulate the operation of that model.

Data Poisoning (Datenvergiftung) Source

‘Adversarial attacks in which an adversary interferes with a model during its training stage, such as by inserting malicious training data (data poisoning) or modifying the training process itself (model poisoning).’

Concept Comparisons

Attack / FailureLlm Security

Prompt Injection vs Data Poisoning

A side-by-side comparison of Prompt Injection and Data Poisoning. Understand how runtime instruction manipulation differs from attacks on training data or the training process.

Attack / FailureLlm Security

Data Poisoning vs Adversarial Learning

A side-by-side comparison of Data Poisoning and Adversarial Learning. Understand how the terms differ, when each applies, and what the distinction means for AI governance, system design, or assurance evidence.

Related Terms

Performance (Leistung)AngriffsflächeData Augmentation (Datenerweiterung)TrainingsdatenBöswilliger ZweckGenerative Model (generatives Modell)