Threat modelling, attack modelling or attack surface mapping for the software takes in the vulnerability and threat types particular to AI models: poisoning of training data, malicious code or unwanted content in inputs and outputs, denial of service from adversarial prompts, supply chain attack, unauthorised information disclosure, theft of weights and misconfigured data pipelines. The organisation may refresh the risk model periodically for later model versions and derivatives after release, and may check during modelling that the model does not sit in a critical path where significant security decisions are made with no human involved.
The graph holds this control, the 0 it maps to, and the evidence behind each claim, over MCP and REST.