Governments should legally mandate immediate reporting of AI security breaches