Defining AI Models and AI Systems: A Framework to Resolve the Boundary Problem
Authors: Yuanyuan Sun, Timothy Parker, Lara Gierschmann, Sana Shams, Teo Canmetin, Mathieu Duteil, Rokas Gipiškis, Ze Shen Chin
Organizations: AI Governance Exchange · Independent · University of British Columbia · Oxford Internet Institute, University of Oxford · AI Standards Lab · Vilnius University · Oxford Martin AI Governance Initiative
Emerging AI regulations assign distinct obligations to different actors along the AI value chain (e.g., the EU AI Act distinguishes providers and deployers for both AI models and AI systems), yet the foundational terms "AI model" and "AI system" lack clear, consistent definitions. Through a systematic review of 896 academic papers and a manual review of over 80 regulatory, standards, and technical or policy documents, we analyze existing definitions from multiple conceptual perspectives. We then trace definitional lineages and paradigm shifts over time, finding that most standards and regulatory definitions derive from the OECD's frameworks, which evolved in ways that compounded rather than resolved conceptual ambiguities. The ambiguity of the boundary between an AI model and an AI system creates practical difficulties in determining obligations for different actors, and raises questions on whether certain modifications performed are specific to the model as opposed to the non-model system components. We propose conceptual definitions grounded in the nature of models and systems and the relationship between them, then develop operational definitions for contemporary neural network-based machine-learning AI: models consist of trained parameters and architecture, while systems consist of the model plus additional components including an interface for processing inputs and outputs. Finally, we discuss implications for regulatory implementation and examine how our definitions contribute to resolving ambiguities in allocating responsibilities across the AI value chain, in both theoretical scenarios and case studies involving real-world incidents.
Figures & tables
Figure 1: Overview of conceptual perspectives for definitions of AI model and AI system .
Figure 2: Definitional lineage of terms related to AI models and AI systems
Figure 3: OECD (2019b) (top) versus Russell and Norvig (2009) (bottom)
AI system / Intelligent Agent
“An AI system is a machine-based system that is capable of influencing the Environment by making recommendations, predictions or decisions for a given set of Objectives. It does so by utilising machine and/or human-based inputs/data to: i) perceive real and/or virtual environments; ii) abstract such perceptions into models manually or automatically; and iii) use Model Interpretations to formulate options for outcomes.” ( OECD, 2019b , p. 7)
AI Operational Logic (p. 6)
“The key power of an AI system resides in its Operational Logic, which, for a given set of objectives and based on input data from Sensors, provides output for the Actuators — as recommendations, predictions or decisions — that are capable of influencing the state of the Environment.” ( OECD, 2019b , p. 6)
Model (p. 6-7)
“A Model is an actionable representation of all or part of the external environment of an AI system that describes the environment’s structure and/or dynamics. The model represents the core of an AI system. A model can be based on data and/or expert knowledge, by humans and/or by automated tools like machine learning algorithms. Model Interpretation is the process of deriving an outcome from a model.” ( OECD, 2019b , p. 7) “Model (a data object constructed by the Model Building process)” ( OECD, 2019b , p. 6)
Table 1: Key definitions related to AI models and AI systems from OECD (2019b)
Russell & Norvig’s Term
OECD (2019b) Equivalent
Key Difference in Taxonomy
Primary Object Type
Agent / Rational Agent / Intelligent Agent
AI System / Intelligent Agent
The definitions are almost identical.
OECD: A mixture of real, digital, and abstract objects.
Agent Function
AI Operational Logic
Near-identical definitions (mapping inputs to outputs)
Abstract
Agent Program
(Most analogous to) Model
Agent program is concrete software; OECD model is “data object”
Digital
World Model (of a model-based agent)
A specific type of Model
Russell and Norvig (2009) : specific to subset of agents (model-based); OECD: implied for all systems
Digital or Abstract
Table 2: Relationship and differences between Russell and Norvig (2009) term and OECD (2019b)
Figure 4: OECD (2022) ’s conceptualisation of the term AI model
Aspect
OECD (2019b)
OECD (2022)
Consequences
Core Definition
An actionable representation of all or part of the external environment of an AI system.
An actionable representation of all or part of the external context or environment of an AI system.
Slight expansion from “environment” to “context.”
Relationship with Environment
The model […] describes the environment’s structure and/or dynamics.
The model represents, describes, and interacts with real or virtual environments.
Introduces the incorrect idea that a model can “interact.”
Relation to System
Clearly a component within the system, representing its environment.
Blurred; diagrammatically equated with the system’s core logic, definitionally remains a representation.
Fundamental tension between the visual framework and its textual definitions.
Table 3: Differences and similarities between terminologies in OECD (2019b) and OECD (2022)
Figure 5: OECD (2024) ’s conceptual diagram
Concept
Russell and Norvig (2009)
OECD (2019b)
OECD (2022) (Text)
OECD (2022) (Image)
OECD (2024)
World model of a model-based agent
Yes
Yes
Yes
Sometimes
Sometimes
Representation of (part of) the external environment
Yes
Yes
Yes
Sometimes
Sometimes
Core component of an AI system
Sometimes (model-based only)
Yes
Yes
Yes
Yes
Abstract mapping between system input and output
No (Agent Function)
No (AI Operational Logic)
Sometimes
Yes
Yes
Policy/Q-function of a model-free agent
No
No
No
Yes
Yes
Table 4: Conceptions of AI model across different frameworks
Figure 6: AI system containing an AI model
Figure 7: AI system containing an AI model with other non-model components
Core component of AI system to make inferences
Parameters change during the build phase
Usually remain fixed after deployment
Part of AI model ?
Model weights
Yes
Yes
Yes
Yes
Input prompt
Yes
Not parameters
Not parameters
No
Activations
Not a component - they are dynamic intermediate computation results
Not parameters
Not parameters
No
System prompt
Not a component - it’s an input of the model
Not parameters
Not parameters
No
RAG
Not a component - it’s a technique; but the associated knowledge base used by RAG can be a core component of the system
Not parameters
Not parameters
No
Temperature setting
Not a component - it’s a setting
Not parameters
Not parameters
No
Table 5: Assessment of whether an element is part of an AI model according to the three candidate criteria from OECD (2024)
Component of AI system
Trained parameters
Part of AI model ?
Model weights
Yes
Yes
Yes
Input prompt
No
No
No
Activations
Not really
No
No
System prompt
No
No
No
RAG
No for the technique itself; yes for the associated knowledge base, if implemented as part of the system
No
No
Temperature setting
No
No
No
Table 6: Assessment of whether an element is part of an AI model according to our candidate criteria
Conceptual definition
Summarised operational definition for contemporary AI
AI model
A computational construct that makes inferences from inputs to produce outputs and forms (or is intended to form) the reasoning or knowledge core of an AI system.
Trained parameters and their architecture, understood as the computational specification necessary to use them for inference.
AI system
A machine-based system that makes inferences from inputs from physical or virtual environments to produce outputs that influence physical or virtual environments.
One or more AI models, an interface for receiving inputs from and delivering outputs to an environment, and the configuration connecting these and any additional components.
Table 7: Our proposed conceptual and summarised operational definitions of AI model and AI system
Appendix figures & tables6 assets
Supplementary material from the paper’s appendix.
Appendix
Criterion
Keywords
Population
AI model and AI system Synonyms: Machine learning system, autonomous system, automated system, algorithmic system Machine learning model, foundation model, algorithmic model, predictive model AI application, machine learning application, AI solution, machine learning solution, AI services
Find different definitions, clarity on differences
Context
Computer science, engineering, social sciences, decision sciences, business, management and accounting, arts and humanities, multidisciplinary, psychology
Appendix
Table 8: SLR Keywords from PICOC Criteria
Category
Selection Criteria
Period
From 2012 to present
Language
English
Relevance to research question
The article provides a definition of AI model , AI system , or a concept synonymous with these terms
Type of source
Articles, reviews and conference papers
Appendix
Table 9: SLR Inclusion and Exclusion Criteria
Figure 8: SLR Search String
Figure 9: PRISMA Flow Diagram of the SLR
Authors
Term defined
Source of definition
Mühlhoff and Ruschemeier [2025]
Machine learning model Machine learning model
Authors themselves Mühlhoff and Ruschemeier [2024]
Kallab et al. [2024]
Machine learning model Machine learning model
Authors themselves Janiesch et al. [2021]
Fernández-Llorca et al. [2025]
AI model Machine learning model
ISO/IEC [2022e] ISO/IEC [2022e]
Lu et al. [2024]
AI model
Authors themselves
Hupont et al. [2024]
AI model AI model
ISO/IEC [2022e] OECD [2024]
Park et al. [2024]
AI model
Authors themselves
Appendix
Table 10: SLR Findings for AI model and machine learning model
Authors
Term defined
Source of definition
Ojanen [2025]
AI system
European Parliament and Council of the European Union [2024]
Müller et al. [2025]
AI system
Authors themselves
Fernández-Llorca et al. [2025]
AI system AI system AI system AI system
OECD [2019a] European Commission [2021] European Parliament [2023] European Parliament and Council of the European Union [2024, Article 3(1)]
Lu et al. [2024]
AI model
Authors themselves
Hupont et al. [2024]
AI system AI system
Authors themselves OECD [2019a]
Buiten [2024]
AI system AI system AI system
Authors themselves Buiten et al. [2023] European Parliament and Council of the European Union [2024]
Appendix
Table 11: SLR Findings for AI system and “machine learning system”
Artificial intelligence governance studies how laws, institutions, and technical controls shape AI development, deployment, accountability, and social effects.
The European AI Act is the first comprehensive regulation of artificial intelligence (AI), setting out extensive obligations, particularly for so-called high-risk and general-purpose AI systems. A key distinguishing feature of AI systems under the AI Act is the capability to infer. Since the AI Act does not clearly define what inference is, there is a gray area for certain data-driven systems. A specific example is credit scoring systems, which are listed by Annex III of the AI Act. At the same time, however, these are often implemented using statistical models for which it is unclear whether they have the capability to infer and thus fall under the AI definition of the AI Act at all. Motivated by statistical learning theory, this work develops a framework for grading different levels of the capability to infer. Based on the AI Act and the Commission Guidelines on the definition of an artificial intelligence system, we analyze which levels constitute sufficient capability to infer within the meaning of the AI Act and where further regulatory clarity is needed. We illustrate the framework by creating two realistic credit scoring workflows and show whether and where inference occurs in them. Our analysis illustrates that not only individual models but the entire data processing workflow must be considered. It also shows that the involvement of human experts during development can have significant influence on the capability to infer. Code can be found at https://github.com/fraunhofer-iais/inference-framework-creditscorecards.
Maximilian Poretschkin, Tabea Naeven
Fraunhofer Institute for Intelligent Analysis and Information Systems (IAIS) University of Bonn Lamarr Institute for Machine Learning and Artificial Intelligence Sankt Augustin, Bonn, Germany · Fraunhofer Institute for Intelligent Analysis and Information Systems (IAIS) Sankt Augustin, Germany
The rapid proliferation of artificial intelligence (AI) technologies has led to a dynamic regulatory landscape, where legislative frameworks strive to keep pace with technical advancements. As AI paradigms shift towards greater autonomy, specifically in the form of agentic AI, it becomes increasingly challenging to precisely articulate regulatory stipulations. This challenge is even more acute in the domains of security and privacy, where the capabilities of autonomous agents often blur traditional legal and technical boundaries. This paper reviews the evolving European Union (EU) AI regulatory provisions via analyzing 24 relevant documents published between 2024 and 2025. From this review, we provide a clarification of critical definitions. We deconstruct the regulatory interpretations of security, privacy, and agentic AI, distinguishing them from closely related concepts to resolve ambiguity. We synthesize the reviewed documents to articulate the current state of regulatory provisions targeting different types of AI, particularly those related to security and privacy aspects. We analyze and reflect on the existing provisions in the regulatory dimension to better align security and privacy obligations with AI and agentic behaviors. These insights serve to inform policymakers, developers, and researchers on the compliance and AI governance in the society with increasing algorithmic agencies.
Shiliang Zhang, Sabita Maharjan
Department of Informatics, University of Oslo, Norway
Due to ambiguity in the wording of the EU AI Act, we examine the question of to what extent frontier biological foundation models such as ESM3 are subject to obligations for general-purpose AI models with systemic risk under the EU AI Act. In this paper, we map ESM3 to the biorisk chain, and conclude that it would be desirable if the providers of ESM3 and similar biological models were subject to these obligations, which would require them to assess and mitigate dual-use risks from their models. We then perform an analysis, comparing the attributes of ESM3 to the classification criteria in the AI Act and the supporting material. We conclude that at this time, ESM3 does not appear to be meaningfully regulated by the Act. We then propose remedies to correct the situation.
Taro Qureshi, Jacob Griffith, Koen Holtman +3
Northeastern University London · London School of Economics and Political Science · AI Standards Lab +2