https://www.theguardian.com/technology/article/2024/may/20/ai-chatbots-safeguards-can-be-easily-bypassed-say-uk-researchers

All five systems tested were found to be ‘highly vulnerable’ to attempts to elicit harmful responses