Domain 1 Part D: AI Security Program Development and Management

Question 1

What is “AI security”?

Answer 1

Systematic management of risks to ensure AI systems are robust, reliable, and resilient against threats

Question 2

What is the acronym to remember the 8 protocols and practices for an AI security program?

Answer 2

STDDDAMP

Question 3

What are the 8 protocols/practices for an AI security program?

Answer 3

1. Societal adaptation
2. Trust but verify
3. Design AUP
4. Designate AI lead
5. Develop set of AI ethics
6. Adapt cyber program
7. Mandate audits and traceability
8. Perform a cost analysis

Question 4

What is the idea behind “trust but verify”?

Answer 4

Outputs must be constantly validated

Question 5

What are 3 prominent models for AI ethics?

Answer 5

• UNESCO, Recommendation on the Ethics of AI
• IBM, “What are AI Ethics?”
• U.S. Department of Defense

Question 6

What are 8 categories of AI metrics?

Answer 6

1. Accountability
2. Fairness
3. Human well-being
4. Performance
5. Privacy and data governance
6. Robustness and digital security
7. Safety
8. Transparency and explainability

Question 7

What is the mnemonic device to remember the 8 categories of AI metrics?

Answer 7

FASTHRPP

Question 8

As a metric, what does accountability mean?

Answer 8

• Obligation to ensure systems operate in ethical, fair, transparent, and compliant manner
• Ensures outputs traceable to responsible party

Question 9

As a metric, what does fairness mean?

2

Answer 9

Systems should be designed to:

• Avoid bias
• Ensure equitable outcomes

Question 10

As a metric, what does well-being mean?

Answer 10

Enhance human flourishing while respecting human rights and fundamental freedoms

Question 11

As a metric, what does performance mean?

Answer 11

Achieve intended purpose effectively and efficiently

Question 12

As a metric, what does privacy and data governance mean?

Answer 12

Respect privacy rights and ensure secure and ethical management of data

Question 13

As a metric, what does robustness and digital security mean?

Answer 13

Resilient and secure against adversarial threats and errors

Question 14

As a metric, what does safety mean?

Answer 14

Minimize risk to people, property, and the environment

Question 15

As a metric, what does transparency and explainability mean?

Answer 15

Rationale behind outputs should be understandable and accessible

Question 16

What are 5 AI security use cases?

Answer 16

• Log analysis
• APT malware detection
• Behaviorial AI
• Threat intelligence
• Phishing detection

Question 17

What actions does AI-enabled log analysis take?

2

Answer 17

• Capture and pair errors with knowledge base
• Create new error and suggest solution

Question 18

What is an APT?

2

Answer 18

• Sophisticated, targeted long-term cyber attack
• Long-term unauthorized access without detection

Question 19

How does APT malware detection work?

Answer 19

• Performs probabilistic inference on unknown PowerShell commands
• Implicitly learns nonlinear combinations/patterns

Question 20

What does Behavioral AI focus on?

Answer 20

Human behavior to detect anomalies/threats

Question 21

How does AI-enabled threat intelligence work?

2

Answer 21

• Existing tools + GenAI
• Provides insights on new patterns, detection