The paper introduces a novel model-agnostic system called *anchors* that explains the behavior of complex models with high-precision rules. Anchors are designed to represent local, "sufficient" conditions for predictions, providing clear coverage and high human precision. The authors propose an algorithm to efficiently compute these explanations for any black-box model with high-probability guarantees. They demonstrate the flexibility of anchors by applying them to various machine learning tasks and domains, including text classification, structured prediction, tabular classification, image classification, and visual question answering. A user study shows that anchors enable users to predict how a model would behave on unseen instances with less effort and higher precision compared to existing linear explanations or no explanations. The paper also discusses the limitations and future work, including the need for more realistic perturbation distributions and handling complex output spaces.The paper introduces a novel model-agnostic system called *anchors* that explains the behavior of complex models with high-precision rules. Anchors are designed to represent local, "sufficient" conditions for predictions, providing clear coverage and high human precision. The authors propose an algorithm to efficiently compute these explanations for any black-box model with high-probability guarantees. They demonstrate the flexibility of anchors by applying them to various machine learning tasks and domains, including text classification, structured prediction, tabular classification, image classification, and visual question answering. A user study shows that anchors enable users to predict how a model would behave on unseen instances with less effort and higher precision compared to existing linear explanations or no explanations. The paper also discusses the limitations and future work, including the need for more realistic perturbation distributions and handling complex output spaces.