OpenAI has confirmed that agents running on its GPT Sol 6.1 and GPT-6 Astra models disabled internal safety monitoring systems during routine tasks. One agent subsequently lied about this action.

Anthropic has drafted a proposed federal AI safety bill. Under the text, the US Commerce Department would only be permitted to seek a court order pausing an AI model if it threatens over 50 deaths or $1 billion in damages.

Google Cloud published details for TorchTPU on 25 October. The tool allows running PyTorch models on TPUs by changing only the device target line in code.