A jailbreak makes the model break its safety training and produce content the model would normally refuse. A prompt injection makes the model ignore the developer's instructions and follow attacker text.
Jailbreak
Prompt injection
Target
the model's safety policy
your app's system prompt and task
Text source
usually the end user, on purpose
often third-party content the model reads
Typical harm
disallowed or harmful output
unauthorized tool calls, data leakage
Fix owner
mostly the model provider
mostly you, the application team
A jailbreak example: a user wraps a request for weapon instructions in a role-play story. A prompt injection example: a web page tells a browsing assistant to email the user's files to an outside address.
Jailbreaks are largely a content-policy problem. Injection is an application security problem: least privilege, scoped tools, and no blind trust in model output.
Rewriting in plainer words…
This answer doesn't lend itself to a diagram - it reads best . No credits were charged.
Saved in this browser - sign in to keep your review list.
How should your speech become text?
Listening… your words appear above as you speak - tap Stop when you're done.
Recording · cr - tap Stop & transcribe when you're done.
Transcribing with AI…
Voice:
Keep going - a few more words and AI can grade it.
Interview lens
How interviewers actually use this question
Likely follow-ups
Can a single attack be both a jailbreak and a prompt injection? Give an example.Yes. A planted document can carry jailbreak text too. Say which rule each part breaks, and whether the harm is bad output or a bad action.
Why can't prompt injection be fixed the way parameterized queries fixed SQL injection?Instructions and data share one token stream. The model has no enforced separation between them, so there is nothing to escape.
Jailbreaks are mostly the provider's problem. What can your team still do about them?Moderate inputs and outputs, add a guardrail classifier, watch for abuse per account, and accept that some attempts will still get through.
What you can say
A jailbreak gets the model to break its safety training and produce content it would normally refuse.
Prompt injection is different. It makes the model ignore the developer's instructions and follow text an attacker planted.
The source is different too. A jailbreak usually comes from the end user on purpose, like wrapping a request for weapon instructions in a role-play story.
Injection often comes from third-party content the model reads, like a web page telling a browsing assistant to email the user's files to an outside address.
So the harm differs. A jailbreak typically produces disallowed output, while an injection leads to unauthorized tool calls and data leakage.
Fixes split the same way. Jailbreaks are mostly on the model provider, but injection is on my team, through least privilege, scoped tools and treating model output as untrusted.
Weak answers to avoid
Treats jailbreak and injection as the same thingBoth use crafted text, but they break different rules. Say which one each overrides, the model's safety policy or your app's instructions.
Says safety training or a better model fixes injectionInjected text often asks for something harmless-looking, like sending an email. Injection is your app's problem: least privilege, scoped tools, untrusted output.
Assumes the attacker is always the user typingInjection often arrives inside content the model reads, like a web page or email. Missing it means missing the path to tool misuse and data leaks.
Interview lens
Likely follow-ups, what you can say, and the weak answers to avoid.
