MITRE ATLAS Framework for AI Threats Quiz

1. In MITRE ATLAS, what does 'evasion' refer to?

Avoiding detection of attacks on AI systems

Crafting inputs designed to fool a model at inference time

Hiding malware from antivirus software

Preventing data exfiltration

2. What is the primary value of using MITRE ATLAS in an organization's AI security strategy?

It eliminates all AI security risks automatically

It provides a standardized framework for identifying, understanding, and defending against AI threats

It replaces the need for traditional cybersecurity measures

It guarantees that AI models will never be compromised

3. In MITRE ATLAS, 'persistence' tactics aim to achieve which goal?

Maintaining long-term access to compromised AI systems

Improving model accuracy over time

Reducing computational overhead

Automating model retraining

4. Which of these is a recommended defense for protecting against adversarial examples?

Adversarial training and input perturbation detection

Removing all user inputs from models

Using only shallow neural networks

Disabling model inference entirely

5. MITRE ATLAS emphasizes that AI security requires collaboration between which groups?

Only security teams

Security, ML engineers, and domain experts

Only AI researchers

Government agencies exclusively

6. What is a backdoor attack in the context of AI security?

An unauthorized remote access to AI infrastructure

Inserting hidden triggers into models that cause specific malicious behavior

Stealing API credentials from developers

Physically accessing data center facilities

7. Which defense mitigation is most effective against supply chain attacks on AI models?

Using only open-source models

Verifying model provenance and integrity through cryptographic signatures

Running models offline

Increasing model training epochs

8. In MITRE ATLAS, 'impact' tactics are primarily concerned with which outcome?

Detecting malicious activity in AI systems

Achieving the adversary's final objective through AI compromise

Securing training data

Improving model robustness

9. What is the primary difference between model stealing and model extraction?

Model stealing targets training data; extraction targets model weights

Both terms are synonymous in AI security

Model extraction is a subset of model stealing techniques

Model stealing involves physical theft; extraction is digital

10. Which of the following is NOT typically a phase in the MITRE ATLAS attack lifecycle?

Reconnaissance

Resource Development

Decryption and Obfuscation

Execution

11. What does MITRE ATLAS stand for in the context of AI security?

Adversarial Threat Language and Exploitation System

Adversarial Tactics, Techniques, and Common Knowledge for AI Security

Automated Threat Learning and Security Framework

Advanced Tactical Learning and AI Safeguards

12. Which defense strategy directly counters model extraction attacks?

Input validation on training data

Rate limiting and monitoring API access to models

Increasing model complexity

Removing all user feedback mechanisms

13. What is model poisoning in the context of MITRE ATLAS?

Injecting malicious data during model training to degrade performance

Stealing trained model weights from servers

Analyzing model decisions to find vulnerabilities

Replacing legitimate models with fake ones

14. Adversarial examples are primarily designed to cause which outcome in AI models?

Increased training time

Model misclassification or incorrect predictions

Improved model accuracy

Reduced computational requirements

15. Which of these is an example of a reconnaissance tactic in AI attacks?

Gathering information about target AI models and their training data

Deploying malware on AI servers

Encrypting model weights

Implementing access controls

16. What is a 'technique' in MITRE ATLAS?

A specific method or procedure used to execute an adversarial tactic

A defensive countermeasure for AI systems

A training algorithm for machine learning models

A hardware component protecting AI infrastructure

17. Which tactic involves an adversary attempting to degrade AI model performance through poisoned data?

Reconnaissance

Resource Development

Initial Access

Execution

18. In MITRE ATLAS, what is a 'tactic' in the context of AI attacks?

A specific malware variant targeting AI systems

A high-level adversary goal or objective in attacking AI

A network vulnerability in machine learning infrastructure

A tool used to exploit AI models

19. MITRE ATLAS is structured similarly to which well-known security framework?

NIST Cybersecurity Framework

ISO 27001

MITRE ATT&CK

CIS Controls

20. Which of the following is a primary goal of the MITRE ATLAS framework?

Eliminate all AI-related security risks

Provide a common language for AI adversarial techniques and defenses

Replace traditional cybersecurity frameworks

Limit AI development in organizations

SecAI+ MITRE ATLAS Framework for AI Threats Quiz