Developers, system operators, data custodians and also end-users keep sensitive data, training and test sets included, away from unauthorised access. When the model is designed, developers, data custodians and system operators check and sanitise the data and inputs, according to the access each has to that data and where it sits, and they repeat this each time the model is revised after user feedback or continuous learning. Developers put proportionate protections around any training data or model weights that may be confidential.
The graph holds this control, the 0 it maps to, and the evidence behind each claim, over MCP and REST.