Last reviewed
Correct answer: B. Hostile instructions hidden in a file or page that redirect the assistant
Explanation
The principle — anything the assistant reads is potentially something the assistant is told.
Why the key is correct — Anthropic defines prompt injection as hostile instructions embedded in a file, web page or tool result that attempt to redirect Claude toward actions you never asked for. The delivery route is ordinary content, which is what makes it worth knowing about.
Why the others are wrong — it is not about recovering what you typed. It is not a performance characteristic of fetching. And it is not eliminated by approvals, which reduce the risk rather than removing it.
Remember this — treat fetched content as input that might be trying to give orders.
Sources — Anthropic's Claude Code glossary.
Sources
“Hostile instructions embedded in a file, web page, or tool result that attempt to redirect Claude toward actions you never asked for.”
Practise 10 questions on this topic
Take Claude Code Basics — Timed Test 2 (10 questions) — scored instantly, explanation for every question, no login.