Which of the following best explains why AI output could be inaccurate?
Prompt injections exploit runtime interpretation of user-supplied input to manipulate AI output without requiring access to training data or model parameters. An attacker crafts malicious prompts that cause the model to generate inaccurate, misleading, or unintended responses. Model poisoning, by contrast, requires access to training data and the training pipeline—making prompt injections the more realistic and prevalent cause of AI output inaccuracy in deployed systems.
Community Discussion