language model alignment

Researchers discovered that crafting harmful prompts into poetry can bypass the safety guardrails of major AI systems, exposing a critical…