jailbreak

Die Kunst ist eine Tochter der Freiheit. – Friedrich Schiller. One of the studies, »Adversarial Poetry as a Universal Single-Turn Jailbreak Mechanism in Large Language Models« by Bisconti et al., shows that this freedom becomes a problem for many AI models. As soon as queries are made in poetic form, even modern systems respond with significantly less restraint. The researchers found that verse-like “adversarial poetry” prompts significantly weaken the security technology of various models and are effective solely through stylistic camouflage. ...

Cisco’s study “Death by a Thousand Prompts: Open Model Vulnerability Analysis” shows that so‑called multi‑turn jailbreaks against open AI models (Open-Weight-LLMs) succeed in up to 92% of cases, revealing serious security vulnerabilities in their architecture. The text was automatically translated from German into English. The German quotations were also translated in sense.