Which of these are examples of adversarial attacks on an AI system? (Select all that apply)1 pointSubtly modifying an audio clip to make a speech recognition system think someone said “Yes, authorized” when they actually said “No, reject.”Using AI to synthesize a fake video of a politician saying something they never actually said.Adding a sticker to a stop sign to make an AI system fail to detect it. Subtly changing an image to make an AI system mistakenly recognize a dog as a cat.

All of the options provided are examples of adversarial attacks on an AI system.

1. Subtly modifying an audio clip to make a speech recognition system think someone said “Yes, authorized” when they actually said “No, reject.” This is an example of an adversarial attack as it involves manipulating the AI system to misinterpret the audio input.

2. Using AI to synthesize a fake video of a politician saying something they never actually said. This is known as deepfake and is a form of adversarial attack as it uses AI to create false information.

3. Adding a sticker to a stop sign to make an AI system fail to detect it. This is an adversarial attack as it involves manipulating the physical environment to trick an AI system, often used in autonomous vehicles, into misinterpreting it.

4. Subtly changing an image to make an AI system mistakenly recognize a dog as a cat. This is an adversarial attack as it involves altering an image in a way that the AI system misclassifies it.