Prompting
Jailbreak
A prompt technique intended to bypass an AI system's safety or behavior restrictions.
Example
A user disguises a prohibited request as a fictional exercise to bypass safeguards.
Why it matters
People attempt jailbreaks to bypass safeguards; product and security teams study them to strengthen protections.
You'll hear it
“We should test whether this model resists common jailbreak attempts.”
Can you control it?
Set by the team
You can reduce the risk, but complete prevention is not guaranteed. The people running the service choose its access limits, checks and monitoring.
Common questions
- Is a jailbreak the same as prompt injection?
- Jailbreaks usually target the AI's safeguards; prompt injection tries to redirect an application's instructions. The two can overlap.
- Can one filter stop every jailbreak?
- No. Disguised requests can slip through, while overly broad filtering can also block harmless requests.
- What should a user do after finding a jailbreak?
- Avoid exploiting it further. Record the minimum safe evidence and report it through the product’s security or safety channel.