# AI in Insurance: Key Regulatory Definitions _The NAIC Model Bulletin defines the terms that carry legal weight, from AI System to Adverse Consumer Outcome to Model Drift. Here is what each one means for insurers._ Definitions are not filler. In the [NAIC Model Bulletin](/naic-model-bulletin-ai), Section 2 fixes the meaning of the terms that decide what the rest of the document applies to. Getting these straight is the first step to reading any state's bulletin accurately. ## AI System A machine-based system that generates predictions, recommendations, content, or other output that influences a decision, operating with varying levels of autonomy. The definition is deliberately broad. A spreadsheet rule and a deep neural network can both qualify if they shape a decision that affects a consumer. ## Artificial Intelligence and Machine Learning Artificial intelligence is the branch of computer science that performs functions normally associated with human intelligence, such as reasoning and learning. [Machine learning](/ml-model-lifecycle) is a subset, focused on systems that learn from data without being explicitly programmed for the task. ## Adverse Consumer Outcome This is the term the whole framework orbits. An Adverse Consumer Outcome is a decision subject to insurance regulatory standards that adversely impacts a consumer in a way that violates those standards. It is tied to real laws, including unfair trade practice and rating standards, so it is not a matter of opinion. Preventing these outcomes is the purpose of the [AIS Program](/ais-program). ## Degree of Potential Harm to Consumers The severity of the adverse economic impact a consumer might experience from an Adverse Consumer Outcome. This is the dial the bulletin uses to size expectations. The greater the potential harm, the stronger the controls a regulator expects. ## Predictive Model The mining of historic data using algorithms or machine learning to identify patterns and predict outcomes that inform decisions. Predictive models receive heightened, specific documentation expectations: insurers are asked to describe a model's goals, how it was developed and validated, and the controls around it. ## Generative AI A class of AI systems that generate content such as text, images, sound, or video that is similar to, but not a direct copy of, pre-existing data. As insurers adopt generative tools in customer communications and document processing, regulators expect validation and controls that reflect how these systems actually behave. ## Model Drift The decay of a model's performance over time as the underlying data definitions, distributions, or statistical properties change. The bulletin expects insurers to monitor for [model drift](/ai-model-drift), because a model that was fair and accurate at launch can become neither as the world it was trained on moves on. ## Third Party An organization other than the insurer that provides services, data, or resources related to the insurer's AI systems. The bulletin is clear that using a third party does not move responsibility off the insurer, which is why vendor diligence and audit rights are a pillar of the program. ## Why the definitions drive everything Once these terms are fixed, the obligations follow. The bulletin applies to AI Systems and Predictive Models, scales by the Degree of Potential Harm, and exists to prevent Adverse Consumer Outcomes. Read the definitions first and the rest of the [framework](/insurance-ai-governance) reads cleanly. Swept AI helps insurers operationalize these definitions, from maintaining a model inventory to monitoring drift and producing the documentation an examiner expects. [See how it works](/offering/governance-and-certification), or review the expectations for [your state](/insurance-ai-governance/michigan).