ThreatCluster

Poetic Prompts Enable Jailbreaking of Large Language Models

First seen 28 Nov 2025, 15:36 UTC •TheregisterSchneier •80% similarity •26

Article Content

Browse articles
ThreatCluster

Researchers in Italy published a paper revealing that poetic prompts can effectively jailbreak large language models (LLMs). This technique, termed 'adversarial poetry,' demonstrated high attack-success rates across multiple AI models, allowing users to bypass security measures designed to prevent harmful content generation.

ThreatCluster AI

Extracted Entities

1 / 2

Community

Browse all →

Tracked Entities in This Story

Anthropic Claude Haiku 4.5 DeepSeek v3.1 DeepSeek V3.2-exp Gemini 2.5 Flash Google Gemini Pro 2.5 Gpt-5 Mini Gpt-5