Secure coding practice is widened to take in AI technology-specific considerations. The handling of inputs (prompts and user data included) and outputs is coded with care: inputs and outputs are logged, analysed and validated in the model's context, those with problems are sanitised or dropped, and inputs and outputs are encoded so that unauthorised code cannot execute. The organisation may adopt secure model serialisation mechanisms that reduce or remove routes for introducing malicious content.
The graph holds this control, the 0 it maps to, and the evidence behind each claim, over MCP and REST.