Cybersecurity
Security Researchers Use Poetry To Jailbreak AI Models
When prompts were presented in poetic rather than prose form, attack success rates increased from 8% to 43%, on average — a fivefold increase.
Source: Dark Reading
