NIST and Trustworthy AI: The Lab That Never Fines Anyone

NIST's AI framework binds nobody, and laws, hospital buyers and accreditors are making it the yardstick for whether an AI system was built with care. The AI Risk Management Framework from NIST, the federal measurement laboratory, is the closest thing the United States has to an agreed description of how a careful AI maker works. Nothing NIST writes binds anybody until a law, a regulator, a buyer or an accreditor adopts it, and that is how its encryption and security standards came to protect nearly every phone, bank and federal system. The AI framework, released in January 2023 and still at version 1.0, covers a harm those older standards never had to: a model that is secure and patched and still gives wrong answers, or works worse for one group of people than another. It sorts the work into four functions, GOVERN, MAP, MEASURE and MANAGE, with 19 categories and 72 outcomes, and names seven traits of trustworthy AI, with valid and reliable as the base. Its Generative AI Profile lists twelve risks, among them confident false answers, privacy leaks and people trusting a system more than they should. Nobody can be certified against it; the certifiable standard is ISO/IEC 42001. Laws point to it unevenly. Texas names the Generative AI Profile as a defense, Montana requires one class of deployer to consider the framework, and Colorado repealed its defense before it took effect. Healthcare feels it most: federal buyers and hospitals ask AI vendors for testing, bias and monitoring evidence, and the Joint Commission asks hospitals for their vendors' documented limits. Health apps, wearables and AI companion chatbots answer to the FTC, the FDA and state laws that never cite NIST and still ask for what it describes. A working program keeps ten records, from an AI inventory and model cards to a bias measurement, a risk register and an audit trail of what a person decided. MARA, the Collective's clinical assistant in development, is being built to produce that evidence. Article ART-HLC-022. Originally published at https://www.maxresearchcollective.com/articles/nist-and-trustworthy-ai.html