For Individuals
For Companies
Events
About
Contact
LOGIN

Skip to content

LLM Security: Defending Applications Against Injection, Leakage and Model Abuse

Training Data Poisoning and Backdoor Risk

Quiz: Training Data Poisoning and Backdoor Risk

Model Denial of Service and Cost Exhaustion Attacks

Quiz: Model Denial of Service and Cost Exhaustion Attacks

Output Handling: Injection Into Downstream Systems and Browsers

Quiz: Output Handling: Injection Into Downstream Systems and Browsers

Authentication, Authorisation and Multi Tenant Isolation

Quiz: Authentication, Authorisation and Multi Tenant Isolation

Secure Retrieval: Search That Respects Access Control

Quiz: Secure Retrieval: Search That Respects Access Control

Security Testing, Logging and Detection for AI Features

Quiz: Security Testing, Logging and Detection for AI Features

Final exam: LLM Security: Defending Applications Against Injection, Leakage and Model Abuse

2 of 2

Previous Lesson

Quiz: Training Data Poisoning and Backdoor Risk

LLM Security: Defending Applications Against Injection, Leakage and Model Abuse
Training Data Poisoning and Backdoor Risk
Quiz: Training Data Poisoning and Backdoor Risk

Time limit: 0

Quiz Summary

0 of 5 Questions completed

Questions:

Information

You have already completed the quiz before. Hence you can not start it again.

Quiz is loading…

You must sign in or sign up to start the quiz.

You must first complete the following:

Results

Quiz complete. Results are being recorded.

Results

0 of 5 Questions answered correctly

Your time:

Time has elapsed

You have reached 0 of 0 point(s), (0)

Earned Point(s): 0 of 0, (0)
0 Essay(s) Pending (Possible Point(s): 0)

Categories

Not categorized 0%

Review
Answered
Correct
Incorrect

Question 1 of 5

1. Question

What is one of the most insidious risks to large language models in production environments according to the lesson?

Model overfitting due to insufficient data
Correct Incorrect Correct answer

Training data poisoning through malicious examples
Correct Incorrect Correct answer

Unauthorised access to model parameters
Correct Incorrect Correct answer

Incorrect model architecture design
Correct Incorrect Correct answer

Correct / 1 Points

Incorrect / 1 Points
Question 2 of 5

2. Question

Which of the following best describes how a backdoor attack functions in a healthcare chatbot?

The model is trained to ignore all medical queries
Correct Incorrect Correct answer

The model responds with correct advice only when a trigger phrase is present
Correct Incorrect Correct answer

The model is programmed to respond with incorrect advice when a specific trigger is used
Correct Incorrect Correct answer

The model is designed to delete medical records when triggered
Correct Incorrect Correct answer

Correct / 1 Points

Incorrect / 1 Points
Question 3 of 5

3. Question

What is a key requirement for implementing a successful backdoor attack?

The trigger must be easily detectable by users
Correct Incorrect Correct answer

The malicious response must appear plausible to maintain credibility
Correct Incorrect Correct answer

The trigger must be integrated into the model's architecture directly
Correct Incorrect Correct answer

The model must be retrained with new data every week
Correct Incorrect Correct answer

Correct / 1 Points

Incorrect / 1 Points
Question 4 of 5

4. Question

Which of the following is a recommended approach for detecting data poisoning or backdoor attacks?

Monitoring user feedback for unusual queries
Correct Incorrect Correct answer

Conducting regular testing with suspected trigger patterns
Correct Incorrect Correct answer

Using only automated tools to validate data sources
Correct Incorrect Correct answer

Avoiding any data validation processes to reduce risk
Correct Incorrect Correct answer

Correct / 1 Points

Incorrect / 1 Points
Question 5 of 5

5. Question

What is one effective mitigation strategy against data poisoning threats?

Allowing unlimited access to training datasets
Correct Incorrect Correct answer

Maintaining backup datasets from known good sources
Correct Incorrect Correct answer

Removing all data validation checks to speed up training
Correct Incorrect Correct answer

Releasing models without any monitoring systems
Correct Incorrect Correct answer

Correct / 1 Points

Incorrect / 1 Points

Login

Accessing this course requires a login. Please enter your credentials below!

Username or Email Address

Password

Remember Me

Continue with Facebook

Continue with Google

Lost Your Password?

Register

Don't have an account? Register one!

Register an Account

Ready to Lead the AI Revolution?

Join hundreds of executives transforming their careers with Tesseract Academy.

Explore Individual Learning

© 2026 Tesseract Academy. All Rights Reserved.