That's why we're setting up adversarial benchmarks to test if they are doing the thing they promised not to do, because we totally trust them.