Prompt injection in depth
Comes up in: AI engineering, AI system design
This concept is being written. It gets the full five steps — explanation, predict, interview angle, practice with tests, and review cards. Here is exactly what it will cover, so you can study it elsewhere in the meantime.
What it covers
- direct vs indirect injection (via web pages, emails, documents, tool results)
- why it cannot be fully solved by prompting
- defence in depth: least-privilege tools, separating data from instructions, approval gates, output filtering, sandboxing, monitoring
- the "lethal trifecta" idea (private data + untrusted content + exfiltration channel — cite Simon Willison 2025)
Words you will know
What you will practise
- red-team a demo agent
- fix with defence in depth
Interview questions this answers
- How would you secure an agent that reads emails and can send messages?