WASHINGTON—The Cybersecurity and Infrastructure Security Agency and its G7 partners have released new joint guidance designed to improve transparency and security across artificial intelligence systems and supply chains, as governments and businesses face growing pressure to better understand the risks embedded inside rapidly expanding AI technologies.
The guidance, titled “Software Bill of Materials for AI – Minimum Elements,” outlines recommendations for what organizations should disclose about the models, datasets, software components, infrastructure and dependencies used to build AI systems. The effort was developed through the G7 Cybersecurity Working Group and includes participation from the United States, Canada, France, Germany, Italy, Japan, the United Kingdom and the European Union.
CISA said the guidance builds on the concept of a software bill of materials, or SBOM, often described as an “ingredients list” for software that helps organizations identify vulnerabilities, track dependencies and better manage supply-chain risk. The new AI-focused framework expands that concept to account for the additional complexity of artificial intelligence systems, including model provenance, training data, APIs, infrastructure and runtime behavior.
The guidance comes as regulators, financial institutions and cybersecurity experts increasingly warn that opaque AI supply chains could create significant operational and security risks. Analysts interviewed by CSO Online said the recommendations could give security teams stronger leverage in vendor-risk reviews by pushing AI developers to provide greater visibility into how their systems are built and maintained, though experts also cautioned that disclosure alone does not guarantee trustworthiness or security.
The recommendations are voluntary and not intended to be exhaustive, but officials said they represent a shared international baseline that can evolve alongside AI technology. The framework organizes AI supply-chain information into seven major categories, including metadata, models, datasets, infrastructure, security properties and performance indicators, with the broader goal of improving transparency, traceability and cyber resilience across AI ecosystems.