AIR is a tool that helps users to find limits for their classifier’s robustness, using causal methods. - View it on GitHub