What is LLM Jailbreaking?#
Jailbreaking exploits model alignment vulnerabilities through hypothetical roleplay, hypothetical scenarios, or cipher encodings to elicit restricted answers.
Everyday Analogy#
A persuasive con artist talking a security guard into unlocking a vault door by inventing an emergency cover story.
Practical Applications#
- Red-teaming models prior to enterprise deployment
- Engineering robust input-sanitization defensive wrappers