Analysis
What Is an AI Jailbreak? Bypassing Guardrails • SQ Magazine
Publication date not yet evaluated · Added sqmagazine.co.uk
Page published
Coverage timeline
Single-source analysis — one report is available.
Why it matters
AI jailbreaks bypass model-level safety guardrails, and this reference entry synthesizes how standards bodies define and categorize the technique for defenders building mental models.
An SQ Magazine glossary entry explains what an AI jailbreak is, citing NIST's taxonomy of adversarial machine learning attacks that places jailbreaks in the direct prompting attack family used to bypass model guardrails and refusal behavior. The piece distinguishes jailbreaks from prompt injection, describes failure modes like competing objectives and mismatched generalization, and references OWASP and CAISI framings.