Creating Malicious Software with AI GOAA Practice Questions (Page 5)

Questions 21–25

1. 21expert · hard

   A government agency is considering using AI-generated deepfakes for covert operations to discredit a foreign adversary. The agency's legal team is evaluating the ethical and legal implications. Which consideration is most critical?

   • AThe deepfakes must be reversible in case the operation is exposed.
   • BThe deepfakes must be attributed to a third party to maintain plausible deniability.
   • CThe use of deepfakes may violate international laws and treaties regarding information warfare.
   • DThe deepfakes must be technically convincing enough to achieve the operational objective.
   Reveal Solution Discussion Select an answer first
2. 22application · medium

   A university researcher is developing a deepfake detection algorithm and wants to publish a paper. The researcher uses publicly available deepfake datasets and creates new deepfakes for testing. What is the most important ethical consideration?

   • APublish the deepfake generation code to allow other researchers to reproduce the results.
   • BUse only synthetic faces generated by AI to avoid using real people's likenesses.
   • CObtain approval from the university's ethics board before starting the research.
   • DEnsure that any new deepfakes created for testing use consenting individuals and are not publicly released in a way that could harm them.
   Reveal Solution Discussion Select an answer first
4. 23expert · hard

   A security team is developing an AI-based malware detector. They want to make it robust against adversarial examples. They have a limited budget and need to choose a defense strategy. Which approach provides the most effective defense-in-depth against a variety of adversarial attacks?

   • AUse a single, very deep neural network to increase model capacity and make it harder to fool.
   • BRely on a signature-based antivirus as a secondary layer to catch any adversarial examples that bypass the AI detector.
   • CUse adversarial training by augmenting the training set with adversarial examples generated from multiple attack algorithms.
   • DApply input sanitization by removing any file that has a high entropy score, as adversarial examples often have high entropy.
   Reveal Solution Discussion Select an answer first
5. 24application · medium

   A social media platform is implementing automated detection of deepfake videos to prevent misinformation. The platform's current model has a high false-positive rate, causing legitimate videos to be removed. Which approach best reduces false positives while maintaining detection of manipulated media?

   • AUse a multi-modal detection system that combines visual, audio, and metadata analysis.
   • BLower the detection threshold to catch more deepfakes, accepting more false positives.
   • CRely solely on user reports to flag potential deepfakes.
   • DDisable automated detection and manually review all flagged videos.
   Reveal Solution Discussion Select an answer first
6. 25expert · hard

   A red team is developing an AI-powered malware that must evade a commercial endpoint protection product that uses both signature-based and ML-based detection. The team has access to the product's API for testing. They want to minimize the number of malware samples that are detected while also ensuring the malware remains functional. Which strategy is most effective?

   • AGenerate a large number of mutations using a generative model and test each one against the product's API, keeping only those that are undetected.
   • BUse a known evasion technique such as packing or encryption, which is sufficient to bypass both signature and ML detection.
   • CUse a genetic algorithm to evolve malware variants that are both functional and evasive, using the product's detection results as a fitness function.
   • DCraft a single adversarial example that is guaranteed to evade the ML model based on theoretical knowledge of the model's architecture.
   Reveal Solution Discussion Select an answer first