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

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

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

Time limit: 0

Quiz Summary

0 of 10 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 10 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 10

1. Question

What are the three primary categories of threats targeting language model applications according to the lesson?

Injection attacks, data corruption, and model optimisation
Correct Incorrect Correct answer

Injection attacks, data leakage, and model abuse
Correct Incorrect Correct answer

Data leakage, network intrusion, and user error
Correct Incorrect Correct answer

Model abuse, input validation failure, and access control issues
Correct Incorrect Correct answer

Correct / 1 Points

Incorrect / 1 Points
Question 2 of 10

2. Question

In LLM applications, prompt injection attacks occur when malicious input manipulates the model’s behaviour through carefully crafted prompts.

True
Correct Incorrect Correct answer

False
Correct Incorrect Correct answer

Only in traditional web applications
Correct Incorrect Correct answer

Not relevant to LLMs
Correct Incorrect Correct answer

Correct / 1 Points

Incorrect / 1 Points
Question 3 of 10

3. Question

Direct prompt injection involves inserting malicious instructions directly into the input prompt that reaches the language model.

True
Correct Incorrect Correct answer

False
Correct Incorrect Correct answer

Only in specific models
Correct Incorrect Correct answer

It is a form of data corruption
Correct Incorrect Correct answer

Correct / 1 Points

Incorrect / 1 Points
Question 4 of 10

4. Question

What is a common method used by attackers to exploit document-based injection attacks?

Sending emails with malicious attachments that appear legitimate
Correct Incorrect Correct answer

Creating web pages with hidden JavaScript code
Correct Incorrect Correct answer

Using social engineering to trick users into clicking links
Correct Incorrect Correct answer

Exploiting vulnerabilities in network protocols
Correct Incorrect Correct answer

Correct / 1 Points

Incorrect / 1 Points
Question 5 of 10

5. Question

In which context would sanitisation focus on preventing SQL injection through parameterised queries?

HTML output contexts
Correct Incorrect Correct answer

Database insertion contexts
Correct Incorrect Correct answer

API endpoint contexts
Correct Incorrect Correct answer

File upload contexts
Correct Incorrect Correct answer

Correct / 1 Points

Incorrect / 1 Points
Question 6 of 10

6. Question

LLMs may unintentionally reveal confidential information when processing user inputs that contain sensitive data such as personal identification numbers or financial information.

True
Correct Incorrect Correct answer

False
Correct Incorrect Correct answer

Only in healthcare applications
Correct Incorrect Correct answer

Only when handling banking inquiries
Correct Incorrect Correct answer

Correct / 1 Points

Incorrect / 1 Points
Question 7 of 10

7. Question

System prompts can leak through output responses, debugging information, or through model interrogation techniques that reveal underlying prompt structures.

True
Correct Incorrect Correct answer

False
Correct Incorrect Correct answer

Only through output responses
Correct Incorrect Correct answer

Only through debugging information
Correct Incorrect Correct answer

Correct / 1 Points

Incorrect / 1 Points
Question 8 of 10

8. Question

What is the primary risk of excessive agency in Large Language Models?

Models may fail to respond to user queries accurately
Correct Incorrect Correct answer

Models might access or execute beyond their intended scope
Correct Incorrect Correct answer

Models could become too expensive to maintain
Correct Incorrect Correct answer

Models may not integrate properly with existing systems
Correct Incorrect Correct answer

Correct / 1 Points

Incorrect / 1 Points
Question 9 of 10

9. Question

What do model weights contain that makes them a target for adversaries attempting to extract competitive advantages?

The model's training algorithm
Correct Incorrect Correct answer

Proprietary trading strategies and customer data patterns
Correct Incorrect Correct answer

The model's inference speed
Correct Incorrect Correct answer

The model's user interface design
Correct Incorrect Correct answer

Correct / 1 Points

Incorrect / 1 Points
Question 10 of 10

10. 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

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.