The workload of radiologists has grown drastically in recent years, which may lead to increased fatigue and higher risk of diagnostic oversights. Consequently, the automatic detection of anomalous scans is of high practical relevance. Conventional supervised models do not generalize reliably beyond the anomalies seen during training but covering the vast range of possible pathologies in the training data is not feasible. In other domains, anomaly detection (AD) has proven to be a viable approach for this problem. In thoracic CT, however, AD has been little explored so far with previous work relying on private data with a narrowanomaly scope.We address this gap by curating CT-RATE- AD, a public dataset for anomaly detection (AD) in thoracic CT, built on the public CT-RATE dataset. It comprises 5065 scans with 18 anomaly types in the unhealthy subset.Additionally,we evaluate several establishedAD methods, including training-free, reconstruction-based, and self-supervised ones, on CT-RATE-AD. These methods achieve up to 0.65 AUROC, 0.62 AP, and 16% specificity at 95% sensitivity and serve as baselines for future work. The dataset and implementation are publicly available at https://github.com/franziskaweber/CT-RATE-AD.