AI supply chain risks include backdoors in models or data.

• A. Data drift
• B. Fairness auditing
• C. Hidden backdoors in external datasets or models
• D. Hardware malfunctions

Answer: (C)

Backdoors in AI supply chains are malicious insertions in external components—such as datasets or pre-trained models—that cause the system to behave in a hidden, triggered way. This is exactly what the option describes: concealed backdoors embedded in external datasets or models can be activated by attackers to alter behavior, often without obvious signs during normal operation. Relying on third-party data or models creates a supply chain risk because you’re trusting components you didn’t build yourself, and a compromised item can undermine security, integrity, or safety.

Data drift is about changes in the input data distribution over time, which can reduce accuracy but isn’t about a hidden malicious feature. Fairness auditing examines bias and fairness issues, not covert manipulation. Hardware malfunctions involve failures in physical components and aren’t about tampering with data or software supplied from outside.

Understanding this helps you appreciate why the best answer points to hidden backdoors in external datasets or models, a classic supply chain threat that requires rigorous provenance, integrity checks, and testing of third-party components to mitigate.