Good governance starts with knowing what tools you have. Keep an inventory that records, for each tool, its purpose, the decision it supports, the data it uses, who owns it, who is affected, when it was last tested, how staff can depart from it and how people can appeal. Assign an accountable owner who has authority over the tool's use, not just its technical upkeep. Require a review before any new tool is used and whenever an existing one changes. The NIST framework puts governance first for a reason: without clear roles, testing and fixes fall between offices.
Many tools come from vendors, so contracts are where much of the protection is won or lost. Before signing, ask for documentation of the data used to build the tool, the validation method and results by group, known limits, and how the tool will be tested on Minnesota's population. Build in the right to independent testing and audit, access to the information needed to explain decisions to affected people and to hearing officers, notice and approval before the vendor changes the model or its data, regular performance and equity reporting, prompt correction of errors, and an exit plan that lets DHS keep its data and move away from the tool. Do not accept a claim that the method is too proprietary to explain in an appeal; the Idaho litigation shows the cost of that position.
Minnesota's data practices law follows the work. Under Minnesota Statutes section 13.05, subdivision 11, when a government entity contracts with a private person to perform any of its functions, the data that person creates, collects, receives, stores, uses, maintains or disseminates in performing those functions is subject to the Data Practices Act, and the contractor must comply as if it were a government entity. Contracts must include a notice that this applies. Involve the DHS data practices and privacy staff, procurement and legal counsel early in any tool acquisition.
Accountability also runs outward. Engage people affected, including self-advocates, families, community organizations, counties and Tribal Nations, when deciding whether to use a tool, what it should and should not do, and how its results will be explained; involve the Tribal liaison where Tribal Nations are affected. Publish plain-language summaries of what tools are used and how they are tested. Set criteria in advance for pausing or retiring a tool, such as unexplained disparities, rising error rates, appeals that reveal systematic problems or data changes that invalidate testing. Deciding to stop using a tool is a sign of working governance, not failure.
