MAF paragraphs 3.52 to 3.54: decide whether to offer an opt-out from AI (by default or on request) considering risk of harm, reversibility, availability and cost of alternatives, complexity of parallel systems and feasibility; where there is no opt-out, give recourse such as review of the decision and keep chatbot histories for complaints; provide feedback channels (possibly run by the DPO) that also let people correct inaccurate data, and a route to request review of material AI decisions, with human review where a fully automated decision has a material effect.
The graph holds this control, the 0 it maps to, and the evidence behind each claim, over MCP and REST.