During red-team testing of an AI system used for lending decisions, which technique BEST simulates a data poisoning attack?

The correct answer is D. Corrupting training datasets to manipulate outcomes. Data poisoning is an attack where an adversary manipulates the training dataset to corrupt the model's learned behavior-causing it to produce biased, incorrect, or attacker-controlled outputs at inference time. In a lending context, poisoned training data could cause…

AI Security Assurance and Resilience

Question

Options

AAdding noise to output predictions
BStealing model weights
CInputting encrypted data
DCorrupting training datasets to manipulate outcomes

How the community answered

(41 responses)

A
2% (1)
B
5% (2)
C
2% (1)
D
90% (37)

Explanation

Data poisoning is an attack where an adversary manipulates the training dataset to corrupt the model's learned behavior-causing it to produce biased, incorrect, or attacker-controlled outputs at inference time. In a lending context, poisoned training data could cause discriminatory approvals or denials. Option A (adding noise to outputs) is output manipulation, not poisoning. Option B (stealing model weights) is a model extraction/theft attack. Option C (inputting encrypted data) is irrelevant to poisoning and would likely just cause errors.

Topics

#Data Poisoning#Adversarial AI#Red Teaming#AI Security Attacks

Community Discussion

No community discussion yet for this question.

Full AAISM Practice