In AI contracts agencies treat their data, and improvements to it such as cleaning and labelling, as a critical asset, keeping sufficient rights to federal data and its improvements, preventing vendor lock-in and barring vendors from using federal information to train or improve commercial offerings without express permission. Where practicable agencies track procured AI by documenting known capabilities, limitations and intended use, documenting training data provenance, testing and validating performance and risk measures in real-world conditions, checking for overfitting to known test data, writing terms for continuous improvement and monitoring, and requiring post-award evaluation; and they favour interoperable products to keep the federal AI market competitive.
The graph holds this control, the 0 it maps to, and the evidence behind each claim, over MCP and REST.