Secure AI Model Serving Infrastructure Quiz

1. What is prompt injection in AI security?

Increasing the length of model inputs

Malicious input designed to manipulate model behavior

Using multiple prompts simultaneously

Storing prompts in a database

2. What is the role of incident response planning in AI deployment security?

Only documenting successful deployments

Defining procedures to detect, respond to, and recover from security incidents

Preventing all errors

Eliminating the need for monitoring

3. Which security control is critical for protecting AI model intellectual property?

Publishing model architecture online

Model watermarking and access restrictions

Using small models only

Disabling all monitoring

4. How should organizations handle model bias and fairness in production AI systems?

Ignore fairness concerns

Audit models regularly and implement mitigation strategies

Remove demographic data from inputs

Use only black-box models

5. What is the purpose of API gateways in secure AI model serving infrastructure?

Increase inference speed only

Centralize authentication, rate limiting, and request validation

Store model weights

Reduce model accuracy

6. Which approach best mitigates data poisoning attacks on deployed AI models?

Accepting all user-provided training data

Validating and sanitizing data before model retraining

Disabling model updates

Removing data logging entirely

7. What does zero-trust architecture mean in AI deployment security?

No security measures are needed

Trust all internal network traffic

Verify every access request regardless of source

Use only network firewalls

8. How should AI model serving infrastructure handle data retention for compliance?

Keep all logs indefinitely

Delete logs immediately

Define retention policies based on regulatory requirements

Encrypt logs without deletion

9. What is the primary benefit of containerization in secure AI deployment?

Faster GPU processing

Isolation of model runtime and dependencies from host system

Reduced model accuracy

Automatic model updates

10. Which monitoring capability is essential for detecting anomalous behavior in deployed AI models?

Tracking only successful API requests

Logging inference latency metrics

Monitoring input/output distributions and error rates

Recording user IP addresses only

11. What is the primary security concern when deploying large language models in a cloud environment?

Model inference latency

Unauthorized access to model weights and sensitive input data

GPU memory allocation

Network bandwidth usage

12. How should sensitive credentials for AI deployment be managed?

Store in plaintext configuration files

Use a secrets management system with encryption and access controls

Embed in application source code

Share via email across teams

13. What does differential privacy protect against in machine learning?

Slow model inference

Individual data point identification from model outputs

GPU failures

Network latency

14. Which of the following is a best practice for securing AI model serving endpoints?

Expose all endpoints publicly without rate limiting

Implement rate limiting and request validation

Store API keys in environment variables without encryption

Log all requests with full input/output data

15. What is model extraction in the context of AI security?

Removing sensitive features from training data

Stealing model weights or behavior through API queries

Exporting model predictions to a database

Compressing model size for deployment

16. Which compliance framework is most relevant for AI systems handling personal data in the EU?

HIPAA

GDPR

SOC 2

PCI DSS

17. In secure AI deployment, what is the purpose of model versioning and rollback capabilities?

To reduce inference costs

To quickly revert to safe versions if vulnerabilities are discovered

To increase model accuracy

To eliminate the need for monitoring

18. Which technique helps prevent adversarial attacks on deployed AI models?

Input validation and sanitization

Increasing model size

Removing model documentation

Disabling user logging

19. What does TLS encryption protect in an AI model serving infrastructure?

Model training data only

Data in transit between client and server

Model weights stored on disk

GPU processing speed

20. Which authentication method is most suitable for securing API endpoints serving AI models?

Basic username/password in URL

OAuth 2.0 with JWT tokens

Hardcoded API keys in client code

No authentication; rely on network isolation

SecAI+ Securing AI Model Serving Infrastructure Quiz