Anthropic and OpenAI have called for mandatory safety checks to be put in place as fears intensify that AI can soon be too powerful to control.