Anthropic called on leading AI labs to agree on a shared framework including potential coordinated slowdowns if advanced self-improving AI systems appear.