Anthropic finds long prompts can jailbreak AI safety

Original: Many Shot Jailbreaking

Source: anthropic.com ↗