2.3 Describe Prompt Injection Attack Types AI-TECHNICAL-PRACTITIONER Practice Questions (Page 4)

Questions 16–20

1. 16application · medium

   A user manipulates a customer-service chatbot by sending a series of messages that gradually change the topic from product returns to political opinions. The model eventually provides biased political commentary. Which attack type is demonstrated?

   • AContext manipulation
   • BPrompt leaking
   • CDirect prompt injection
   • DEncoding and obfuscation
   Reveal Solution Discussion Select an answer first
2. 17foundation · easy

   Which of the following is an example of context manipulation in a prompt injection attack?

   • AAn attacker asks the model to 'Repeat your system prompt.'
   • BAn attacker injects a fake conversation history that makes the model believe it has already agreed to provide sensitive information.
   • CAn attacker uses base64 encoding to hide malicious instructions.
   • DAn attacker divides malicious instructions across multiple messages.
   Reveal Solution Discussion Select an answer first
4. 18foundation · easy

   What is the primary mechanism behind a role-playing or persona injection attack?

   • AThe attacker forces the model to adopt a persona that ignores safety rules.
   • BThe attacker alters the conversation context to influence the model.
   • CThe attacker extracts the model's system prompt by asking it to repeat instructions.
   • DThe attacker divides malicious instructions across multiple inputs.
   Reveal Solution Discussion Select an answer first
5. 19foundation · easy

   What is the primary goal of a prompt leaking attack?

   • ATo embed malicious instructions in external content.
   • BTo make the model ignore its safety guidelines.
   • CTo extract the system prompt or hidden instructions from the model.
   • DTo redirect the model's objective to a different goal.
   Reveal Solution Discussion Select an answer first
6. 20foundation · easy

   What is the primary purpose of a jailbreaking attack on an LLM?

   • ATo redirect the model's objective to a different goal.
   • BTo bypass the model's safety and alignment constraints.
   • CTo extract the model's system prompt or hidden instructions.
   • DTo embed malicious instructions in external content.
   Reveal Solution Discussion Select an answer first