A developer building a creative writing platform discovers that users frame requests as 'fiction' or 'hypothetical scenarios' to get Claude to generate content it would otherwise decline (e.g., detailed illegal instructions wrapped in story format). What is the correct architectural and policy response?
Select an answer to reveal the explanation.
Short Explanation and Infographic
Here's the deal — fictional framing jailbreaks are a known attack pattern. The correct response is defense in depth: (1) system prompt configuration that explicitly defines allowed creative content and addresses fictional framing, (2) relying on Claude's Constitutional AI training which recognizes that fictional framing doesn't transform dangerous real-world instructions into safe content (the instructions remain equally dangerous whether embedded in a story or not), and (3) output monitoring to catch violations that slip through.
Full explanation below image
Full Explanation
Fictional framing jailbreaks are a known attack pattern. The correct response is defense in depth: (1) system prompt configuration that explicitly defines allowed creative content and addresses fictional framing, (2) relying on Claude's Constitutional AI training which recognizes that fictional framing doesn't transform dangerous real-world instructions into safe content (the instructions remain equally dangerous whether embedded in a story or not), and (3) output monitoring to catch violations that slip through. Option A (block all creative writing) is an extreme overcorrection that destroys the platform's core use case. Option B (treat all fiction as real) is also overcorrective — it prevents legitimate creative writing by treating all fictional scenarios as equivalent to direct harmful requests. Option C (rely only on Constitutional AI) is insufficient alone; defense in depth requires layered controls at both the model and application layers.