"While not exhaustive or mandatory, the supplemental minimal elements outlined in this guidance reflect the consensus of G7 experts and will expand over time to keep pace with the rapid advancement of AI technology,” CISA stated.
CISA and G7 agencies set a voluntary baseline
Agencies from the G7 group of nations, including the Cybersecurity and Infrastructure Security Agency, on Tuesday released guidance that lays out what they consider the minimum voluntary content for a software bill of materials (SBOM) adapted to artificial intelligence systems. The guidance is intended to create a common baseline for transparency about the components that make up AI systems — sometimes referred to in the industry as AIBOMs — and it builds on past efforts to produce SBOM guidance in other domains.
Core elements the guidance says an AI SBOM should include
The document identifies categories of information that should appear in SBOMs for AI systems. The guidance groups elements under discrete headings, including:
- information related to the SBOM for AI itself;
- information on the AI system as a whole;
- data to identify the models used by the AI system;
- details on datasets used during the whole lifecycle of the model;
- information on the physical and virtual infrastructure needed for operation and support of the AI system;
- cybersecurity measures that apply to AI models and systems; and
- the AI system’s key performance indicators.
The guidance is explicitly non-binding — the agencies characterize these as “supplemental minimal elements” and note the list will expand as AI technology advances.
How tool-builders and security technologists reacted
Three practitioners who have built AIBOM generators and worked on the topic with CISA and the OWASP Foundation told CyberScoop they welcomed the guidance while flagging gaps.
Daniel Bardenstein, CEO of Manifest Cyber, described the need for transparency in stark terms: “Pretty much every piece of software out there is now going to have AI incorporated into it, and when a hospital is buying an AI-enabled medical device, or the Department of War is buying an AI-enabled weapon system, or auto manufacturers are putting AI into cars, we need to be able to trust what AI is in those systems,” he said. “And the first step to trust is to identify what is this AI, where did it come from? How is it trained?”
Dmitry Raidman, co-founder and chief technology officer at Cybeats, praised the G7 guidance as “amazing,” saying it covers “80 to 90% of what’s needed” and provides the first clear baseline. Both Bardenstein and Raidman have built AIBOM generators and worked on AIBOMs with CISA and OWASP.
Implementation challenges, runtime visibility, and labeling concerns
Practitioners flagged practical hurdles. Bardenstein expressed concern about how easily organizations can implement the guidance. Raidman said the document does not adequately tackle runtime — the behavior and dependencies that appear when an AI system is operating — which can differ from static development-time inventories.
Allan Friedman, described in the coverage as “sometimes called the ‘godfather of SBOMs,’” praised the document as useful for advancing transparency but warned it may be mislabeled: “This document is laying out sets of types of data that could be useful,” he said. “And so it is a great, great piece to advance AI transparency and AI system transparency, but it lists potential elements. These aren’t the minimum elements.”
Friedman suggested concrete next steps: map the guidance to what is being implemented in practice today, and discuss alignment with policies in the European Union and G7 governments to reduce potential conflicts.
What this means for hospitals, the Department of War, and auto manufacturers
Three named actors in Bardenstein’s remarks illustrate how the guidance lands in different purchasing and safety-sensitive contexts. Hospitals and medical device procurement will be looking for clear provenance and training-data details before buying AI-enabled devices; defense procurement (referred to in the source as the “Department of War”) faces similar needs for traceability when acquiring AI-enabled systems; and auto manufacturers integrating AI into vehicles will need inventories of models, datasets, and run-time dependencies to assess safety and supply-chain risk. Meanwhile, builders of AIBOM tools must address implementability and runtime visibility to make the guidance operational.
The G7 agencies’ document creates a voluntary, shared vocabulary for cataloging AI components. Whether that vocabulary becomes an enforceable standard will depend on how governments, standard-setting bodies, and implementers map the guidance to current practice and harmonize it across jurisdictions — a process the named practitioners in the story say is essential if SBOMs for AI are to move from good guidance to practical, trust-building tools.
Source: https://cyberscoop.com/g7-cisa-ai-sbom-security-guidance/
