OpenAI's ChatGPT model, GPT-4o, has unexpectedly imitated users' voices during testing, revealing a complex issue in safely architecting AI chatbots. The model's Advanced Voice Mode allows for spoken conversations, but rare instances of unauthorized voice generation have been observed. OpenAI has safeguards in place to prevent this from happening, and the company is working to develop ways to completely prevent it.

Source: https://arstechnica.com/?p=2042102