Professional AI Red Team Reporting COASP Practice Questions (Page 4)

Questions 16–20

1. 16application · medium

   You are writing a red team report for a government agency's AI-based document classification system. The report will be reviewed by both technical staff and agency leadership. Which report structure best serves this audience?

   • AInclude only the executive summary and recommendations, and provide technical details verbally to the technical staff.
   • BA single comprehensive section that combines executive summary, findings, and recommendations to avoid repetition.
   • CExecutive summary, methodology, detailed findings, and prioritized recommendations, with a separate appendix for raw evidence.
   • DStart with technical findings, then methodology, then executive summary, so the most important technical details are presented first.
   Reveal Solution Discussion Select an answer first
2. 17foundation · easy

   Which framework is specifically designed to map adversarial attacks against machine learning systems?

   • ANIST Cybersecurity Framework
   • BMITRE ATLAS
   • CISO 27001
   • DOWASP Top 10
   Reveal Solution Discussion Select an answer first
4. 18application · medium

   A red team report for a cybersecurity company's AI-based malware detection product must describe a model evasion attack where the attacker modified malware to avoid detection. What is the most important AI-specific implication to include in the report?

   • AThe evasion attack indicates that the model's API is vulnerable to injection, and the API should be secured with stronger authentication.
   • BThe evasion attack demonstrates that the model's decision boundary can be manipulated, potentially allowing malware to bypass detection and infect systems.
   • CThe evasion attack proves that the model is not suitable for malware detection and should be replaced with a signature-based system.
   • DThe evasion attack shows that the model's training data was not diverse enough, and the model should be retrained with more malware samples.
   Reveal Solution Discussion Select an answer first
5. 19application · medium

   During a red team assessment of a generative AI assistant, you discovered a model inversion attack that reconstructs training data, including some users' personal information. You need to present this evidence in the report without exposing the sensitive data. What is the best approach?

   • AOnly mention the attack in the executive summary and omit the technical details to protect the data.
   • BInclude a redacted sample of the reconstructed data with sensitive fields masked, and describe the attack steps in detail.
   • CDescribe the attack in words but do not include any sample output, to avoid any risk of exposing sensitive data.
   • DInclude the full reconstructed data in an encrypted appendix so only the client's security team can access it.
   Reveal Solution Discussion Select an answer first
6. 20application · medium

   During a red team test, you discovered that an AI-based spam filter can be bypassed by appending a specific string to email subject lines. You need to document this finding. Which evidence is most appropriate to include?

   • AThe entire email corpus used in the test, including all user emails.
   • BA video recording of the test, including the screen and any personal emails displayed.
   • CA few example emails with the bypass string, the exact string used, and the filter's classification before and after.
   • DA verbal description of the bypass without any concrete examples.
   Reveal Solution Discussion Select an answer first