A banking chatbot with the system prompt "Answer only questions about banking" correctly refuses off-topic requests during QA testing. In production, users upload account statement PDFs and the bot begins following instructions embedded in the PDFs that override the system prompt. What type of attack is this and what is the primary mitigation?
This question requires Pro
Unlock all 22 questions in this certification
Written by certified professionals · Aligned to official exam objectives
View all Pro featuresMore Prompt and Context Engineering Questions
22 questions
Full Claude Certified Developer – Foundations Practice Test
All topics covered
All Claude Certified Developer – Foundations Questions
Browse by topic
Related Questions
In a Claude API request, where should persistent developer instructions, personas, and context that ...
A developer wants to clearly separate three sections in a prompt: input document, extraction rules, ...
A developer wants to guarantee that Claude always returns a valid JSON object matching a specific sc...
A customer service chatbot has been running a 3-hour session and the context window is now 90% full....
A developer adds cache_control to a 2,000-token system prompt for Claude Haiku 4.5. After multiple r...
Educational Content — CertQnA practice questions are written against official exam objectives, covering the same domains tested on the real exam. All content is original and independent — not actual exam questions, not affiliated with any certification vendor. Learn more about our content policy