Make sure the AI system or service performs sufficiently against its requirements, including accuracy of output (reliability); keep performance at level under varied conditions and avoid grossly wrong judgments on unrelated events (robustness); secure controllability so that people can control the AI where needed, including periodic and objective monitoring and response, in proportion to the seriousness and likelihood of rights infringement from its use or unintended behaviour; carry out an appropriate risk analysis and treat the risks (avoid, reduce, transfer or accept); where harm to life, body, property, mind or the environment is possible, organise in advance the measures to take and give stakeholders the related information, stating the measures relevant stakeholders should take and the terms of use; and prepare how to respond when the safety of the system is compromised so that the response can be executed quickly.
The graph holds this control, the 0 it maps to, and the evidence behind each claim, over MCP and REST.