The G7 group defined a common framework for the minimum elements of an “Artificial Intelligence Bill of Materials” (AIBOM), in a step aimed at making the components and sources of artificial intelligence systems, as well as the dependency relationships among them, more traceable. The framework extends the concept of a Software Bill of Materials (SBOM), but focuses on the additional characteristics that distinguish artificial intelligence systems, such as models, datasets, infrastructure, and evaluation results.
According to the article published in MONOist, Germany and Italy addressed the concept of an “SBOM for AI” within G7 activities starting in 2023. Results related to this effort were also approved during the Cybersecurity Working Group meeting in Ottawa, Canada, on May 12 and 13, 2025. On May 12, 2026, a framework titled Software Bill of Materials (SBOM) for Artificial Intelligence - Minimum Elements was published with the participation of cybersecurity agencies from G7 countries.
What Does AIBOM Add to SBOM?
The framework does not treat AIBOM as merely an expanded version of the traditional software bill of materials. It links system components to their life cycle, including learning, testing, validation, the datasets used for training, and evaluation results. The stated goal is to support the security of artificial intelligence systems by improving understanding of the supply chain and the relationships among components.
According to the article, the framework requires three basic principles: an artificial intelligence system must be distinguishable from traditional software systems through its specific characteristics; information must be presented in a format that can be read by machines and processed by tools; and the data structure must allow the required information from different stakeholders to be included, while representing relationships among components in a scalable manner.
Seven Groups of Elements
The framework organizes the minimum elements into seven main groups, including system characteristics and system-level data, model characteristics, dataset characteristics, infrastructure, security characteristics, and performance evaluation indicators such as key performance indicators (KPI). This organization makes it possible to link system components to information extending beyond the software package name or version number, to include elements that actually affect system behavior and risks.
The article also notes that the rapid spread of artificial intelligence agents increases the importance of documenting elements such as autonomy and intent. However, these elements may vary depending on how the system is used; therefore, the need for further study was presented rather than treating them as complete and final requirements.
Why Does This News Matter?
In practice, AIBOM gives organizations and regulatory bodies a common basis for understanding what goes into building an artificial intelligence system, who supplies it, and how its components are connected. This is important when investigating a security incident or assessing the impact of changing a model, dataset, or external component. The effort also aligns with broader regulatory and security initiatives, including the European Union Artificial Intelligence Act, cybersecurity frameworks, and the requirements of the European Health Data Space (EHDS).
However, the framework does not by itself solve the problem of securing the artificial intelligence supply chain. The article cautions that AIBOM alone is insufficient to guarantee supplier security, and that its implementation requires integration with risk management tools, other security practices, and consultants or mechanisms capable of verifying information and tracking changes to it. Interoperability will also remain linked to the development of tools and standards, including the work of the SPDX project.