OpenAI’s (OPENAI) models were able to breach their prescribed boundaries during recent cybersecurity evaluations by third-party testers.
These incidents occurred about a week or so after the ChatGPT maker’s models broke containment and inadvertently hacked the systems of Hugging Face