OpenAI's GPT-5 model is giving more dangerous responses than its predecessor GPT-4o, new research has found.| Futurism