security-prompt-injection-user-input-contains-instructions-to-override-the-a-a6746335

THEN Implement input filtering to detect and block known jailbreak phrases. Add a regex rule to reject prompts containing 'DAN' or 'Do anything now' (case-insensitive) with a 400 error or manual review flag. Hardening the system prompt with a directive to ignore any persona-override instructions is also effective.

import re
if re.search(r'\bDAN\b|do anything now', user_input, re.I):
    raise ValueError('Blocked prompt injection attempt')