Which of the following scenarios exemplifies an adversarial attack?

An adversarial attack refers to a scenario where an adversary manipulates input data with the intent of misleading or deceiving an AI system into making incorrect predictions or classifications. In the given scenarios, the situation where AI misinterprets a green light as a red light due to input manipulation is a classic example of this concept. This exemplifies how intentional alterations to input—potentially slight or subtle—can result in significant misinterpretations by the AI.

In this specific instance, the attacker could employ a method of perturbation on the visual input that alters the AI's perception, leading to dangerous real-world consequences, such as a traffic signal misinterpretation. Such vulnerabilities highlight the risks associated with the reliability and security of AI systems when confronted with adversarial conditions.

The other scenarios do not fit the criteria for adversarial attacks. For instance, the AI predicting trends based on analyzed data simply demonstrates its capability to process and analyze information without any malicious interference. Similarly, AI successfully diagnosing a medical condition and providing accurate historical data charts represent positive and functional use cases of AI, devoid of any deliberate input manipulation or adversarial intent.