The Stealthiest AI Attack

Data poisoning attacks manipulate AI model performance by injecting malicious samples into training data, with 'clean-label attacks' being especially hard to detect as they alter features while maintaining correct labels.

Sources

Open the full topic