To avoid harm from provision or use not envisaged at development, set clear policies and guidance on how the AI can be used safely; and when post-training a pre-trained model, select the pre-trained model appropriately, checking whether its licence allows commercial use, its pre-training data, the specifications needed for training and running it, and its authenticity.
The graph holds this control, the 0 it maps to, and the evidence behind each claim, over MCP and REST.