A task new to the Profile: model weights and configuration parameter data are protected from unauthorised access and modification. Weights and parameters are kept apart from training, test, fine-tuning and alignment data; their integrity, and for closed models their confidentiality, is continuously monitored; least privilege restricts access to weights, parameters and services during development; and extra risk-proportionate measures are specified and applied around weights, such as encryption, hashing, signing, authorisation by more than one person and air-gapped environments.
The graph holds this control, the 0 it maps to, and the evidence behind each claim, over MCP and REST.