MAF paragraphs 3.31 to 3.35: test robustness against foreseeable erroneous input through scenario-based testing and, for higher-risk or public-facing features, adversarial testing; be aware of the risks of continual learning; keep an internal policy and process for regular model tuning as behaviour, objectives, risks or values change; test in a way that reflects the dynamic production environment so solutions generalise and fail gracefully, and actively monitor, review and tune after deployment.
The graph holds this control, the 0 it maps to, and the evidence behind each claim, over MCP and REST.