An artifact store is the storage layer in an MLOps platform where workflow outputs are saved, such as trained models, evaluation reports, datasets, logs, and other run artifacts. In systems like MLflow, the tracking database may hold metadata, while the artifact store holds the actual files.
It matters in cyber security because those files are often trusted by other tools and teams. If an attacker can tamper with an artifact store, they may replace a model, poison outputs, steal sensitive training data, or hide evidence of malicious activity. Defenders should treat it like a protected production repository: restrict network access, require authentication, encrypt data at rest, and log every read or write. In incident response, unexpected changes to artifacts or unusual download activity can be a sign of compromise.