A recommendation still needs a name on it
When software produces a recommendation, accountability is not a label on the model. It is a record of who is responsible, what was measured, and how the thing will be watched after it is in use, as two public frameworks already say.
Escherbridge
When software produces a recommendation, accountability is not a label on the model. It is a record of who is responsible, what data went in, whether the output was checked against a stated goal, and how the thing will be watched after it is in use. Two public frameworks already say this in plain terms. Neither of them is a certification, and this essay does not claim that PrimusNeo, NEOS, or AZOA implements either one.
The National Institute of Standards and Technology published the Artificial Intelligence Risk Management Framework (AI RMF 1.0) in January 2023, as NIST AI 100-1. "As directed by the National Artificial Intelligence Initiative Act of 2020 (P.L. 116-283), the goal of the AI RMF is to offer a resource to the organizations designing, developing, deploying, or using AI systems to help manage the many risks of AI." "The Framework is intended to be voluntary, rights-preserving, non-sector-specific, and use-case agnostic." It is written to equip "organizations and individuals – referred to here as AI actors – with approaches that increase the trustworthiness of AI systems." Characteristics of trustworthy AI systems, in the document's list, "include valid and reliable, safe, secure and resilient, accountable and transparent, explainable and interpretable, privacy enhanced, and fair with their harmful biases managed." The Core describes four functions — GOVERN, MAP, MEASURE, and MANAGE — and GOVERN applies "to all stages of organizations' AI risk management processes and procedures." "Actions do not constitute a checklist, nor are they necessarily an ordered set of steps." "Risk management should be continuous, timely, and performed throughout the AI system lifecycle dimensions."
The U.S. Government Accountability Office published its own framework in June 2021, Artificial Intelligence: An Accountability Framework for Federal Agencies and Other Entities (GAO-21-519SP). "This framework is organized around four complementary principles, which address governance, data, performance, and monitoring." It is written for "federal agencies and other entities involved in the design, development, deployment, and continuous monitoring of AI systems," and it names the oversight problem directly: "AI systems pose unique challenges to such oversight because their inputs and operations are not always visible." Practice 1.1 asks the reader to "define clear goals and objectives for the AI system to ensure intended outcomes are achieved." Practice 1.2 asks them to "define clear roles, responsibilities, and delegation of authority for the AI system to ensure effective operations, timely corrections, and sustained oversight." This essay does not treat PrimusNeo as a federal system, and it does not treat the GAO document as a certificate anyone here holds.
The same four NIST functions have a published companion for generative systems. In July 2024, NIST issued Artificial Intelligence Risk Management Framework: Generative Artificial Intelligence Profile (NIST AI 600-1), "a cross-sectoral profile of and companion resource for the AI Risk Management Framework (AI RMF 1.0) for Generative AI." The AI RMF, the profile repeats, "is intended for voluntary use and to improve the ability of organizations to incorporate trustworthiness considerations into the design, development, use, and evaluation of AI products, services, and systems." A profile, in that document's wording, "is an implementation of the AI RMF functions, categories, and subcategories for a specific setting, application, or technology." The useful fact here is a narrow one: the same four functions have a voluntary companion, dated July 2024, for generative systems. This page does not summarize a risk catalogue it has not quoted.
A recommendation that cannot name a person, a goal, and a way to check it later is not ready to run a place. That is the same habit as writing the test before you build, applied after the system is in use. If a later paragraph needs a door into the stack, the already-public ones will do: agreements and permissions, and a control that fails closed rather than waving through what it cannot verify. ArdaNova is the workspace that keeps a record of work and of decisions on agreed terms. None of those pages is a claim that the frameworks above have been implemented. This is not legal advice.
Sources
- NIST, Artificial Intelligence Risk Management Framework (AI RMF 1.0), NIST AI 100-1, January 2023
- NIST AI RMF 1.0 DOI
- U.S. GAO, Artificial Intelligence: An Accountability Framework for Federal Agencies and Other Entities, GAO-21-519SP, June 2021
- NIST, Artificial Intelligence Risk Management Framework: Generative Artificial Intelligence Profile, NIST AI 600-1, July 2024
- NIST AI 600-1 DOI
Related
- Field guide: AI accountability (NIST AI RMF & GAO)Primary framework quotes across GOVERN, MAP, MEASURE, and MANAGE.
- Field guide: What to log when software recommendsThe five fields—goal, input, validation check, owner, and monitoring.
- Pilot purgatory: write the test before you buildThe same habit, earlier. Write the test before anyone builds.
- ACT is not a vote
- Fail closed, or it isn't a control