“Anthropic's Responsible Scaling Policy (RSP) considers that computer use capabilities could enable future powerful models to execute risks at the ASL-3 and ASL-4 levels.”